experiments/STT
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge branch 'rebrand' onto main

Browse Source
This commit is contained in:
Kelly Davis 2021-03-07 14:42:44 +01:00
commit 742b44dd2c
281 changed files with 1874 additions and 1771 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
.gitmodules vendored
View File

@ -1,10 +1,10 @@
[submodule "doc/examples"]
path = doc/examples
url = https://github.com/mozilla/DeepSpeech-examples.git
url = https://github.com/coqui-ai/STT-examples.git
branch = master
[submodule "tensorflow"]
path = tensorflow
url = https://github.com/mozilla/tensorflow.git
url = https://github.com/coqui-ai/tensorflow.git
[submodule "kenlm"]
path = kenlm
url = https://github.com/kpu/kenlm

0
.taskcluster.yml → .taskcluster.yml.disabled
View File

3
BIBLIOGRAPHY.md
View File

@ -1,5 +1,4 @@
This file contains a list of papers in chronological order that have been published
using DeepSpeech.
This file contains a list of papers in chronological order that have been published using 🐸STT.
To appear
==========

139
CODE_OF_CONDUCT.md
View File

@ -1,15 +1,132 @@
# Community Participation Guidelines
# Contributor Covenant Code of Conduct
This repository is governed by Mozilla's code of conduct and etiquette guidelines.
For more details, please read the
[Mozilla Community Participation Guidelines](https://www.mozilla.org/about/governance/policies/participation/).
## Our Pledge
## How to Report
For more information on how to report violations of the Community Participation Guidelines, please read our '[How to Report](https://www.mozilla.org/about/governance/policies/participation/reporting/)' page.
We as members, contributors, and leaders pledge to make participation in our
community a harassment-free experience for everyone, regardless of age, body
size, visible or invisible disability, ethnicity, sex characteristics, gender
identity and expression, level of experience, education, socio-economic status,
nationality, personal appearance, race, caste, color, religion, or sexual identity
and orientation.
<!--
## Project Specific Etiquette
We pledge to act and interact in ways that contribute to an open, welcoming,
diverse, inclusive, and healthy community.
In some cases, there will be additional project etiquette i.e.: (https://bugzilla.mozilla.org/page.cgi?id=etiquette.html).
Please update for your project.
-->
## Our Standards
Examples of behavior that contributes to a positive environment for our
community include:
* Demonstrating empathy and kindness toward other people
* Being respectful of differing opinions, viewpoints, and experiences
* Giving and gracefully accepting constructive feedback
* Accepting responsibility and apologizing to those affected by our mistakes,
and learning from the experience
* Focusing on what is best not just for us as individuals, but for the
overall community
Examples of unacceptable behavior include:
* The use of sexualized language or imagery, and sexual attention or
advances of any kind
* Trolling, insulting or derogatory comments, and personal or political attacks
* Public or private harassment
* Publishing others' private information, such as a physical or email
address, without their explicit permission
* Other conduct which could reasonably be considered inappropriate in a
professional setting
## Enforcement Responsibilities
Community leaders are responsible for clarifying and enforcing our standards of
acceptable behavior and will take appropriate and fair corrective action in
response to any behavior that they deem inappropriate, threatening, offensive,
or harmful.
Community leaders have the right and responsibility to remove, edit, or reject
comments, commits, code, wiki edits, issues, and other contributions that are
not aligned to this Code of Conduct, and will communicate reasons for moderation
decisions when appropriate.
## Scope
This Code of Conduct applies within all community spaces, and also applies when
an individual is officially representing the community in public spaces.
Examples of representing our community include using an official e-mail address,
posting via an official social media account, or acting as an appointed
representative at an online or offline event.
## Enforcement
Instances of abusive, harassing, or otherwise unacceptable behavior may be
reported to the community leaders responsible for enforcement by emailing
[coc-report@coqui.ai](mailto:coc-report@coqui.ai).
All complaints will be reviewed and investigated promptly and fairly.
All community leaders are obligated to respect the privacy and security of the
reporter of any incident.
## Enforcement Guidelines
Community leaders will follow these Community Impact Guidelines in determining
the consequences for any action they deem in violation of this Code of Conduct:
### 1. Correction
**Community Impact**: Use of inappropriate language or other behavior deemed
unprofessional or unwelcome in the community.
**Consequence**: A private, written warning from community leaders, providing
clarity around the nature of the violation and an explanation of why the
behavior was inappropriate. A public apology may be requested.
### 2. Warning
**Community Impact**: A violation through a single incident or series
of actions.
**Consequence**: A warning with consequences for continued behavior. No
interaction with the people involved, including unsolicited interaction with
those enforcing the Code of Conduct, for a specified period of time. This
includes avoiding interactions in community spaces as well as external channels
like social media. Violating these terms may lead to a temporary or
permanent ban.
### 3. Temporary Ban
**Community Impact**: A serious violation of community standards, including
sustained inappropriate behavior.
**Consequence**: A temporary ban from any sort of interaction or public
communication with the community for a specified period of time. No public or
private interaction with the people involved, including unsolicited interaction
with those enforcing the Code of Conduct, is allowed during this period.
Violating these terms may lead to a permanent ban.
### 4. Permanent Ban
**Community Impact**: Demonstrating a pattern of violation of community
standards, including sustained inappropriate behavior, harassment of an
individual, or aggression toward or disparagement of classes of individuals.
**Consequence**: A permanent ban from any sort of public interaction within
the community.
## Attribution
This Code of Conduct is adapted from the [Contributor Covenant][homepage],
version 2.0, available at
[https://www.contributor-covenant.org/version/2/0/code_of_conduct.html][v2.0].
Community Impact Guidelines were inspired by
[Mozilla's code of conduct enforcement ladder][Mozilla CoC].
For answers to common questions about this code of conduct, see the FAQ at
[https://www.contributor-covenant.org/faq][FAQ]. Translations are available
at [https://www.contributor-covenant.org/translations][translations].
[homepage]: https://www.contributor-covenant.org
[v2.0]: https://www.contributor-covenant.org/version/2/0/code_of_conduct.html
[Mozilla CoC]: https://github.com/mozilla/diversity
[FAQ]: https://www.contributor-covenant.org/faq
[translations]: https://www.contributor-covenant.org/translations

16
CODE_OWNERS.rst
View File

@ -1,7 +1,7 @@
DeepSpeech code owners / governance system
==========================================
Coqui STT code owners / governance system
=========================================
DeepSpeech is run under a governance system inspired (and partially copied from) by the `Mozilla module ownership system <https://www.mozilla.org/about/governance/policies/module-ownership/>`_. The project is roughly divided into modules, and each module has its own owners, which are responsible for reviewing pull requests and deciding on technical direction for their modules. Module ownership authority is given to people who have worked extensively on areas of the project.
🐸STT is run under a governance system inspired (and partially copied from) by the `Mozilla module ownership system <https://www.mozilla.org/about/governance/policies/module-ownership/>`_. The project is roughly divided into modules, and each module has its own owners, which are responsible for reviewing pull requests and deciding on technical direction for their modules. Module ownership authority is given to people who have worked extensively on areas of the project.
Module owners also have the authority of naming other module owners or appointing module peers, which are people with authority to review pull requests in that module. They can also sub-divide their module into sub-modules with their own owners.
@ -46,7 +46,7 @@ Testing & CI
Native inference client
-----------------------
Everything that goes into libdeepspeech.so and is not specifically covered in another area fits here.
Everything that goes into libstt.so and is not specifically covered in another area fits here.
- Alexandre Lissy (@lissyx)
- Reuben Morais (@reuben)
@ -110,7 +110,7 @@ Documentation
- Alexandre Lissy (@lissyx)
- Reuben Morais (@reuben)
Third party bindings
--------------------
Hosted externally and owned by the individual authors. See the `list of third-party bindings <https://deepspeech.readthedocs.io/en/master/USING.html#third-party-bindings>`_ for more info.
.. Third party bindings
--------------------
Hosted externally and owned by the individual authors. See the `list of third-party bindings <https://stt.readthedocs.io/en/latest/ USING.html#third-party-bindings>`_ for more info.

25
CONTRIBUTING.rst
View File

@ -1,37 +1,32 @@
Contribution guidelines
=======================
Welcome to the DeepSpeech project! We are excited to see your interest, and appreciate your support!
Welcome to the 🐸STT project! We are excited to see your interest, and appreciate your support!
This repository is governed by Mozilla's code of conduct and etiquette guidelines. For more details, please read the `Mozilla Community Participation Guidelines <https://www.mozilla.org/about/governance/policies/participation/>`_.
This repository is governed by the Contributor Covenant Code of Conduct. For more details, see the `CODE_OF_CONDUCT.md <CODE_OF_CONDUCT.md>`_.
How to Make a Good Pull Request
-------------------------------
Here's some guidelines on how to make a good PR to DeepSpeech.
Here's some guidelines on how to make a good PR to 🐸STT.
Bug-fix PR
^^^^^^^^^^
You've found a bug and you were able to squash it! Great job! Please write a short but clear commit message describing the bug, and how you fixed it. This makes review much easier. Also, please name your branch something related to the bug-fix.
Documentation PR
^^^^^^^^^^^^^^^^
If you're just making updates or changes to the documentation, there's no need to run all of DeepSpeech's tests for Continuous Integration (i.e. Taskcluster tests). In this case, at the end of your short but clear commit message, you should add **X-DeepSpeech: NOBUILD**. This will trigger the CI tests to skip your PR, saving both time and compute.
New Feature PR
^^^^^^^^^^^^^^
You've made some core changes to DeepSpeech, and you would like to share them back with the community -- great! First things first: if you're planning to add a feature (not just fix a bug or docs) let the DeepSpeech team know ahead of time and get some feedback early. A quick check-in with the team can save time during code-review, and also ensure that your new feature fits into the project.
You've made some core changes to 🐸STT, and you would like to share them back with the community -- great! First things first: if you're planning to add a feature (not just fix a bug or docs) let the 🐸STT team know ahead of time and get some feedback early. A quick check-in with the team can save time during code-review, and also ensure that your new feature fits into the project.
The DeepSpeech codebase is made of many connected parts. There is Python code for training DeepSpeech, core C++ code for running inference on trained models, and multiple language bindings to the C++ core so you can use DeepSpeech in your favorite language.
The 🐸STT codebase is made of many connected parts. There is Python code for training 🐸STT, core C++ code for running inference on trained models, and multiple language bindings to the C++ core so you can use 🐸STT in your favorite language.
Whenever you add a new feature to DeepSpeech and what to contribute that feature back to the project, here are some things to keep in mind:
Whenever you add a new feature to 🐸STT and what to contribute that feature back to the project, here are some things to keep in mind:
1. You've made changes to the core C++ code. Core changes can have downstream effects on all parts of the DeepSpeech project, so keep that in mind. You should minimally also make necessary changes to the C client (i.e. **args.h** and **client.cc**). The bindings for Python, Java, and Javascript are SWIG generated, and in the best-case scenario you won't have to worry about them. However, if you've added a whole new feature, you may need to make custom tweaks to those bindings, because SWIG may not automagically work with your new feature, especially if you've exposed new arguments. The bindings for .NET and Swift are not generated automatically. It would be best if you also made the necessary manual changes to these bindings as well. It is best to communicate with the core DeepSpeech team and come to an understanding of where you will likely need to work with the bindings. They can't predict all the bugs you will run into, but they will have a good idea of how to plan for some obvious challenges.
1. You've made changes to the core C++ code. Core changes can have downstream effects on all parts of the 🐸STT project, so keep that in mind. You should minimally also make necessary changes to the C client (i.e. **args.h** and **client.cc**). The bindings for Python, Java, and Javascript are SWIG generated, and in the best-case scenario you won't have to worry about them. However, if you've added a whole new feature, you may need to make custom tweaks to those bindings, because SWIG may not automagically work with your new feature, especially if you've exposed new arguments. The bindings for .NET and Swift are not generated automatically. It would be best if you also made the necessary manual changes to these bindings as well. It is best to communicate with the core 🐸STT team and come to an understanding of where you will likely need to work with the bindings. They can't predict all the bugs you will run into, but they will have a good idea of how to plan for some obvious challenges.
2. You've made changes to the Python code. Make sure you run a linter (described below).
3. Make sure your new feature doesn't regress the project. If you've added a significant feature or amount of code, you want to be sure your new feature doesn't create performance issues. For example, if you've made a change to the DeepSpeech decoder, you should know that inference performance doesn't drop in terms of latency, accuracy, or memory usage. Unless you're proposing a new decoding algorithm, you probably don't have to worry about affecting accuracy. However, it's very possible you've affected latency or memory usage. You should run local performance tests to make sure no bugs have crept in. There are lots of tools to check latency and memory usage, and you should use what is most comfortable for you and gets the job done. If you're on Linux, you might find [[perf](https://perf.wiki.kernel.org/index.php/Main_Page)] to be a useful tool. You can use sample WAV files for testing which are provided in the `DeepSpeech/data/` directory.
3. Make sure your new feature doesn't regress the project. If you've added a significant feature or amount of code, you want to be sure your new feature doesn't create performance issues. For example, if you've made a change to the 🐸STT decoder, you should know that inference performance doesn't drop in terms of latency, accuracy, or memory usage. Unless you're proposing a new decoding algorithm, you probably don't have to worry about affecting accuracy. However, it's very possible you've affected latency or memory usage. You should run local performance tests to make sure no bugs have crept in. There are lots of tools to check latency and memory usage, and you should use what is most comfortable for you and gets the job done. If you're on Linux, you might find [[perf](https://perf.wiki.kernel.org/index.php/Main_Page)] to be a useful tool. You can use sample WAV files for testing which are provided in the `STT/data/` directory.
Requesting review on your PR
----------------------------
@ -47,9 +42,9 @@ Before making a Pull Request for Python code changes, check your changes for bas
.. code-block:: bash
pip install pylint cardboardlint
cardboardlinter --refspec master
cardboardlinter --refspec main
This will compare the code against master and run the linter on all the changes. We plan to introduce more linter checks (e.g. for C++) in the future. To run it automatically as a git pre-commit hook, do the following:
This will compare the code against the main branch and run the linter on all the changes. We plan to introduce more linter checks (e.g. for C++) in the future. To run it automatically as a git pre-commit hook, do the following:
.. code-block:: bash

34
Dockerfile.build.tmpl
View File

@ -3,8 +3,8 @@
# Need devel version cause we need /usr/include/cudnn.h
FROM nvidia/cuda:10.1-cudnn7-devel-ubuntu18.04
ENV DEEPSPEECH_REPO=#DEEPSPEECH_REPO#
ENV DEEPSPEECH_SHA=#DEEPSPEECH_SHA#
ENV STT_REPO=#STT_REPO#
ENV STT_SHA=#STT_SHA#
# >> START Install base software
@ -113,15 +113,15 @@ RUN echo "build --spawn_strategy=standalone --genrule_strategy=standalone" \
WORKDIR /
RUN git clone --recursive $DEEPSPEECH_REPO DeepSpeech
WORKDIR /DeepSpeech
RUN git checkout $DEEPSPEECH_SHA
RUN git clone --recursive $STT_REPO STT
WORKDIR /STT
RUN git checkout $STT_SHA
RUN git submodule sync tensorflow/
RUN git submodule update --init tensorflow/
# >> START Build and bind
WORKDIR /DeepSpeech/tensorflow
WORKDIR /STT/tensorflow
# Fix for not found script https://github.com/tensorflow/tensorflow/issues/471
RUN ./configure
@ -132,7 +132,7 @@ RUN ./configure
# passing LD_LIBRARY_PATH is required cause Bazel doesn't pickup it from environment
# Build DeepSpeech
# Build STT
RUN bazel build \
--workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" \
--config=monolithic \
@ -149,22 +149,22 @@ RUN bazel build \
--copt=-msse4.2 \
--copt=-mavx \
--copt=-fvisibility=hidden \
//native_client:libdeepspeech.so \
//native_client:libstt.so \
--verbose_failures \
--action_env=LD_LIBRARY_PATH=${LD_LIBRARY_PATH}
# Copy built libs to /DeepSpeech/native_client
RUN cp bazel-bin/native_client/libdeepspeech.so /DeepSpeech/native_client/
# Copy built libs to /STT/native_client
RUN cp bazel-bin/native_client/libstt.so /STT/native_client/
# Build client.cc and install Python client and decoder bindings
ENV TFDIR /DeepSpeech/tensorflow
ENV TFDIR /STT/tensorflow
RUN nproc
WORKDIR /DeepSpeech/native_client
RUN make NUM_PROCESSES=$(nproc) deepspeech
WORKDIR /STT/native_client
RUN make NUM_PROCESSES=$(nproc) stt
WORKDIR /DeepSpeech
WORKDIR /STT
RUN cd native_client/python && make NUM_PROCESSES=$(nproc) bindings
RUN pip3 install --upgrade native_client/python/dist/*.whl
@ -176,8 +176,8 @@ RUN pip3 install --upgrade native_client/ctcdecode/dist/*.whl
# Allow Python printing utf-8
ENV PYTHONIOENCODING UTF-8
# Build KenLM in /DeepSpeech/native_client/kenlm folder
WORKDIR /DeepSpeech/native_client
# Build KenLM in /STT/native_client/kenlm folder
WORKDIR /STT/native_client
RUN rm -rf kenlm && \
git clone https://github.com/kpu/kenlm && \
cd kenlm && \
@ -188,4 +188,4 @@ RUN rm -rf kenlm && \
make -j $(nproc)
# Done
WORKDIR /DeepSpeech
WORKDIR /STT

18
Dockerfile.train.tmpl
View File

@ -3,8 +3,8 @@
FROM tensorflow/tensorflow:1.15.4-gpu-py3
ENV DEBIAN_FRONTEND=noninteractive
ENV DEEPSPEECH_REPO=#DEEPSPEECH_REPO#
ENV DEEPSPEECH_SHA=#DEEPSPEECH_SHA#
ENV STT_REPO=#STT_REPO#
ENV STT_SHA=#STT_SHA#
RUN apt-get update && apt-get install -y --no-install-recommends \
apt-utils \
@ -20,7 +20,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
unzip \
wget
# We need to remove it because it's breaking deepspeech install later with
# We need to remove it because it's breaking STT install later with
# weird errors about setuptools
RUN apt-get purge -y python3-xdg
@ -31,10 +31,10 @@ RUN apt-get install -y --no-install-recommends libopus0 libsndfile1
RUN rm -rf /var/lib/apt/lists/*
WORKDIR /
RUN git clone $DEEPSPEECH_REPO DeepSpeech
RUN git clone $STT_REPO STT
WORKDIR /DeepSpeech
RUN git checkout $DEEPSPEECH_SHA
WORKDIR /STT
RUN git checkout $STT_SHA
# Build CTC decoder first, to avoid clashes on incompatible versions upgrades
RUN cd native_client/ctcdecode && make NUM_PROCESSES=$(nproc) bindings
@ -43,7 +43,7 @@ RUN pip3 install --upgrade native_client/ctcdecode/dist/*.whl
# Prepare deps
RUN pip3 install --upgrade pip==20.2.2 wheel==0.34.2 setuptools==49.6.0
# Install DeepSpeech
# Install STT
# - No need for the decoder since we did it earlier
# - There is already correct TensorFlow GPU installed on the base image,
# we don't want to break that
@ -54,7 +54,7 @@ RUN python3 util/taskcluster.py --source tensorflow --branch r1.15 \
--artifact convert_graphdef_memmapped_format --target .
# Build KenLM to generate new scorers
WORKDIR /DeepSpeech/native_client
WORKDIR /STT/native_client
RUN rm -rf kenlm && \
git clone https://github.com/kpu/kenlm && \
cd kenlm && \
@ -63,6 +63,6 @@ RUN rm -rf kenlm && \
cd build && \
cmake .. && \
make -j $(nproc)
WORKDIR /DeepSpeech
WORKDIR /STT
RUN ./bin/run-ldc93s1.sh

2
GRAPH_VERSION
View File

@ -1 +1 @@
training/deepspeech_training/GRAPH_VERSION
training/coqui_stt_training/GRAPH_VERSION

2
ISSUE_TEMPLATE.md
View File

@ -1,4 +1,4 @@
For support and discussions, please use our [Discourse forums](https://discourse.mozilla.org/c/deep-speech).
For support and discussions, please use [GitHub Discussions](https://github.com/coqui-ai/STT/discussions).
If you've found a bug, or have a feature request, then please create an issue with the following information:

8
Makefile
View File

@ -1,8 +1,8 @@
DEEPSPEECH_REPO ?= https://github.com/mozilla/DeepSpeech.git
DEEPSPEECH_SHA ?= origin/master
STT_REPO ?= https://github.com/coqui-ai/STT.git
STT_SHA ?= origin/main
Dockerfile%: Dockerfile%.tmpl
sed \
-e "s|#DEEPSPEECH_REPO#|$(DEEPSPEECH_REPO)|g" \
-e "s|#DEEPSPEECH_SHA#|$(DEEPSPEECH_SHA)|g" \
-e "s|#STT_REPO#|$(STT_REPO)|g" \
-e "s|#STT_SHA#|$(STT_SHA)|g" \
< $< > $@

22
README.rst
View File

@ -1,22 +1,20 @@
Project DeepSpeech
==================
.. image:: images/coqui-STT-logo-green.png
:alt: Coqui STT logo
.. image:: https://readthedocs.org/projects/deepspeech/badge/?version=latest
:target: https://deepspeech.readthedocs.io/?badge=latest
.. image:: https://readthedocs.org/projects/stt/badge/?version=latest
:target: https://stt.readthedocs.io/?badge=latest
:alt: Documentation
.. image:: https://img.shields.io/badge/Contributor%20Covenant-2.0-4baaaa.svg
:target: CODE_OF_CONDUCT.md
:alt: Contributor Covenant
.. image:: https://community-tc.services.mozilla.com/api/github/v1/repository/mozilla/DeepSpeech/master/badge.svg
:target: https://community-tc.services.mozilla.com/api/github/v1/repository/mozilla/DeepSpeech/master/latest
:alt: Task Status
**Coqui STT** is an open-source Speech-To-Text engine, using a model trained by machine learning techniques based on `Baidu's Deep Speech research paper <https://arxiv.org/abs/1412.5567>`_. 🐸STT uses Google's `TensorFlow <https://www.tensorflow.org/>`_ to make the implementation easier.
**Documentation** for installation, usage, and training models are available on `stt.readthedocs.io <https://stt.readthedocs.io/>`_.
DeepSpeech is an open-source Speech-To-Text engine, using a model trained by machine learning techniques based on `Baidu's Deep Speech research paper <https://arxiv.org/abs/1412.5567>`_. Project DeepSpeech uses Google's `TensorFlow <https://www.tensorflow.org/>`_ to make the implementation easier.
Documentation for installation, usage, and training models are available on `deepspeech.readthedocs.io <https://deepspeech.readthedocs.io/?badge=latest>`_.
For the latest release, including pre-trained models and checkpoints, `see the latest release on GitHub <https://github.com/mozilla/DeepSpeech/releases/latest>`_.
For the **latest release**, including pre-trained models and checkpoints, `see the latest release on GitHub <https://github.com/coqui-ai/STT/releases/latest>`_.
For contribution guidelines, see `CONTRIBUTING.rst <CONTRIBUTING.rst>`_.

12
RELEASE.rst
View File

@ -1,12 +0,0 @@
Making a (new) release of the codebase
======================================
* Update version in VERSION file, commit
* Open PR, ensure all tests are passing properly
* Merge the PR
* Fetch the new master, tag it with (hopefully) the same version as in VERSION
* Push that to Github
* New build should be triggered and new packages should be made
* TaskCluster should schedule a merge build **including** a "DeepSpeech Packages" task

6
SUPPORT.rst
View File

@ -5,8 +5,8 @@ Contact/Getting Help
There are several ways to contact us or to get help:
#. `Discourse Forums <https://discourse.mozilla.org/c/deep-speech>`_ - The `Deep Speech category on Discourse <https://discourse.mozilla.org/c/deep-speech>`_ is the first place to look. Search for keywords related to your question or problem to see if someone else has run into it already. If you can't find anything relevant there, search on our `issue tracker <https://github.com/mozilla/deepspeech/issues>`_ to see if there is an existing issue about your problem.
#. `GitHub Discussions <https://github.com/coqui-ai/STT/discussions>`_ - `GitHub Discussions <https://github.com/coqui-ai/STT/discussions>`_ is the first place to look. Search for keywords related to your question or problem to see if someone else has run into it already. If you can't find anything relevant there, search on our `issue tracker <https://github.com/coqui-ai/STT/issues>`_ to see if there is an existing issue about your problem.
#. `Matrix chat <https://chat.mozilla.org/#/room/#machinelearning:mozilla.org>`_ - If your question is not addressed by either the `FAQ <https://github.com/mozilla/DeepSpeech/wiki#frequently-asked-questions>`_ or `Discourse Forums <https://discourse.mozilla.org/c/deep-speech>`_\ , you can contact us on the ``#machinelearning`` channel on `Mozilla Matrix <https://chat.mozilla.org/#/room/#machinelearning:mozilla.org>`_\ ; people there can try to answer/help
#. `Matrix chat <https://matrix.to/#/+coqui:matrix.org>`_ - If your question is not addressed on `GitHub Discussions <https://github.com/coqui-ai/STT/discussions>`_\ , you can contact us on the ``#stt:matrix.org`` `channel on Matrix <https://matrix.to/#/#stt:matrix.org?via=matrix.org>`_.
#. `Create a new issue <https://github.com/mozilla/deepspeech/issues>`_ - Finally, if you have a bug report or a feature request that isn't already covered by an existing issue, please open an issue in our repo and fill the appropriate information on your hardware and software setup.
#. `Create a new issue <https://github.com/coqui-ai/STT/issues>`_ - Finally, if you have a bug report or a feature request that isn't already covered by an existing issue, please open an issue in our repo and fill the appropriate information on your hardware and software setup.

2
VERSION
View File

@ -1 +1 @@
training/deepspeech_training/VERSION
training/coqui_stt_training/VERSION

4
bin/compare_samples.py
View File

@ -6,8 +6,8 @@ import sys
import argparse
import numpy as np
from deepspeech_training.util.audio import AUDIO_TYPE_NP, mean_dbfs
from deepspeech_training.util.sample_collections import load_sample
from coqui_stt_training.util.audio import AUDIO_TYPE_NP, mean_dbfs
from coqui_stt_training.util.sample_collections import load_sample
def fail(message):

8
bin/data_set_tool.py
View File

@ -8,20 +8,20 @@ import argparse
import progressbar
from pathlib import Path
from deepspeech_training.util.audio import (
from coqui_stt_training.util.audio import (
AUDIO_TYPE_PCM,
AUDIO_TYPE_OPUS,
AUDIO_TYPE_WAV,
change_audio_types,
)
from deepspeech_training.util.downloader import SIMPLE_BAR
from deepspeech_training.util.sample_collections import (
from coqui_stt_training.util.downloader import SIMPLE_BAR
from coqui_stt_training.util.sample_collections import (
CSVWriter,
DirectSDBWriter,
TarWriter,
samples_from_sources,
)
from deepspeech_training.util.augmentations import (
from coqui_stt_training.util.augmentations import (
parse_augmentations,
apply_sample_augmentations,
SampleAugmentation

2
bin/import_aidatatang.py
View File

@ -5,7 +5,7 @@ import tarfile
import pandas
from deepspeech_training.util.importers import get_importers_parser
from coqui_stt_training.util.importers import get_importers_parser
COLUMN_NAMES = ["wav_filename", "wav_filesize", "transcript"]

2
bin/import_aishell.py
View File

@ -5,7 +5,7 @@ import tarfile
import pandas
from deepspeech_training.util.importers import get_importers_parser
from coqui_stt_training.util.importers import get_importers_parser
COLUMNNAMES = ["wav_filename", "wav_filesize", "transcript"]

6
bin/import_ccpmf.py
View File

@ -30,9 +30,9 @@ except ImportError as ex:
import requests
import json
from deepspeech_training.util.downloader import SIMPLE_BAR, maybe_download
from deepspeech_training.util.helpers import secs_to_hours
from deepspeech_training.util.importers import (
from coqui_stt_training.util.downloader import SIMPLE_BAR, maybe_download
from coqui_stt_training.util.helpers import secs_to_hours
from coqui_stt_training.util.importers import (
get_counter,
get_importers_parser,
get_imported_samples,

8
bin/import_cv.py
View File

@ -10,13 +10,13 @@ from multiprocessing import Pool
import progressbar
import sox
from deepspeech_training.util.downloader import SIMPLE_BAR, maybe_download
from deepspeech_training.util.importers import (
from coqui_stt_training.util.downloader import SIMPLE_BAR, maybe_download
from coqui_stt_training.util.importers import (
get_counter,
get_imported_samples,
print_import_report,
)
from deepspeech_training.util.importers import validate_label_eng as validate_label
from coqui_stt_training.util.importers import validate_label_eng as validate_label
FIELDNAMES = ["wav_filename", "wav_filesize", "transcript"]
SAMPLE_RATE = 16000
@ -35,7 +35,7 @@ def _download_and_preprocess_data(target_dir):
archive_path = maybe_download(ARCHIVE_NAME, target_dir, ARCHIVE_URL)
# Conditionally extract common voice data
_maybe_extract(target_dir, ARCHIVE_DIR_NAME, archive_path)
# Conditionally convert common voice CSV files and mp3 data to DeepSpeech CSVs and wav
# Conditionally convert common voice CSV files and mp3 data to Coqui STT CSVs and wav
_maybe_convert_sets(target_dir, ARCHIVE_DIR_NAME)

10
bin/import_cv2.py
View File

@ -3,7 +3,7 @@
Broadly speaking, this script takes the audio downloaded from Common Voice
for a certain language, in addition to the *.tsv files output by CorporaCreator,
and the script formats the data and transcripts to be in a state usable by
DeepSpeech.py
train.py
Use "python3 import_cv2.py -h" for help
"""
import csv
@ -15,8 +15,8 @@ from multiprocessing import Pool
import progressbar
import sox
from deepspeech_training.util.downloader import SIMPLE_BAR
from deepspeech_training.util.importers import (
from coqui_stt_training.util.downloader import SIMPLE_BAR
from coqui_stt_training.util.importers import (
get_counter,
get_imported_samples,
get_importers_parser,
@ -138,9 +138,9 @@ def _maybe_convert_set(dataset, tsv_dir, audio_dir, filter_obj, space_after_ever
print_import_report(counter, SAMPLE_RATE, MAX_SECS)
output_csv = os.path.join(os.path.abspath(audio_dir), dataset + ".csv")
print("Saving new DeepSpeech-formatted CSV file to: ", output_csv)
print("Saving new Coqui STT-formatted CSV file to: ", output_csv)
with open(output_csv, "w", encoding="utf-8", newline="") as output_csv_file:
print("Writing CSV file for DeepSpeech.py as: ", output_csv)
print("Writing CSV file for train.py as: ", output_csv)
writer = csv.DictWriter(output_csv_file, fieldnames=FIELDNAMES)
writer.writeheader()
bar = progressbar.ProgressBar(max_value=len(rows), widgets=SIMPLE_BAR)

2
bin/import_fisher.py
View File

@ -11,7 +11,7 @@ import librosa
import pandas
import soundfile # <= Has an external dependency on libsndfile
from deepspeech_training.util.importers import validate_label_eng as validate_label
from coqui_stt_training.util.importers import validate_label_eng as validate_label
# Prerequisite: Having the sph2pipe tool in your PATH:
# https://www.ldc.upenn.edu/language-resources/tools/sphere-conversion-tools

2
bin/import_freestmandarin.py
View File

@ -6,7 +6,7 @@ import tarfile
import numpy as np
import pandas
from deepspeech_training.util.importers import get_importers_parser
from coqui_stt_training.util.importers import get_importers_parser
COLUMN_NAMES = ["wav_filename", "wav_filesize", "transcript"]

2
bin/import_gram_vaani.py
View File

@ -12,7 +12,7 @@ import pandas as pd
from sox import Transformer
import swifter
from deepspeech_training.util.importers import get_importers_parser, get_validate_label
from coqui_stt_training.util.importers import get_importers_parser, get_validate_label
__version__ = "0.1.0"
_logger = logging.getLogger(__name__)

2
bin/import_ldc93s1.py
View File

@ -4,7 +4,7 @@ import sys
import pandas
from deepspeech_training.util.downloader import maybe_download
from coqui_stt_training.util.downloader import maybe_download
def _download_and_preprocess_data(data_dir):

2
bin/import_librivox.py
View File

@ -12,7 +12,7 @@ import progressbar
from sox import Transformer
from tensorflow.python.platform import gfile
from deepspeech_training.util.downloader import maybe_download
from coqui_stt_training.util.downloader import maybe_download
SAMPLE_RATE = 16000

4
bin/import_lingua_libre.py
View File

@ -12,8 +12,8 @@ from multiprocessing import Pool
import progressbar
import sox
from deepspeech_training.util.downloader import SIMPLE_BAR, maybe_download
from deepspeech_training.util.importers import (
from coqui_stt_training.util.downloader import SIMPLE_BAR, maybe_download
from coqui_stt_training.util.importers import (
get_counter,
get_imported_samples,
get_importers_parser,

4
bin/import_m-ailabs.py
View File

@ -10,8 +10,8 @@ from multiprocessing import Pool
import progressbar
from deepspeech_training.util.downloader import SIMPLE_BAR, maybe_download
from deepspeech_training.util.importers import (
from coqui_stt_training.util.downloader import SIMPLE_BAR, maybe_download
from coqui_stt_training.util.importers import (
get_counter,
get_imported_samples,
get_importers_parser,

2
bin/import_magicdata.py
View File

@ -6,7 +6,7 @@ import wave
import pandas
from deepspeech_training.util.importers import get_importers_parser
from coqui_stt_training.util.importers import get_importers_parser
COLUMN_NAMES = ["wav_filename", "wav_filesize", "transcript"]

2
bin/import_primewords.py
View File

@ -7,7 +7,7 @@ import tarfile
import numpy as np
import pandas
from deepspeech_training.util.importers import get_importers_parser
from coqui_stt_training.util.importers import get_importers_parser
COLUMN_NAMES = ["wav_filename", "wav_filesize", "transcript"]

4
bin/import_slr57.py
View File

@ -9,8 +9,8 @@ from multiprocessing import Pool
import progressbar
from deepspeech_training.util.downloader import SIMPLE_BAR, maybe_download
from deepspeech_training.util.importers import (
from coqui_stt_training.util.downloader import SIMPLE_BAR, maybe_download
from coqui_stt_training.util.importers import (
get_counter,
get_imported_samples,
get_importers_parser,

4
bin/import_swb.py
View File

@ -1,7 +1,7 @@
#!/usr/bin/env python
# ensure that you have downloaded the LDC dataset LDC97S62 and tar exists in a folder e.g.
# ./data/swb/swb1_LDC97S62.tgz
# from the deepspeech directory run with: ./bin/import_swb.py ./data/swb/
# from the Coqui STT directory run with: ./bin/import_swb.py ./data/swb/
import codecs
import fnmatch
import os
@ -17,7 +17,7 @@ import pandas
import requests
import soundfile # <= Has an external dependency on libsndfile
from deepspeech_training.util.importers import validate_label_eng as validate_label
from coqui_stt_training.util.importers import validate_label_eng as validate_label
# ARCHIVE_NAME refers to ISIP alignments from 01/29/03
ARCHIVE_NAME = "switchboard_word_alignments.tar.gz"

6
bin/import_swc.py
View File

@ -1,6 +1,6 @@
#!/usr/bin/env python
"""
Downloads and prepares (parts of) the "Spoken Wikipedia Corpora" for DeepSpeech.py
Downloads and prepares (parts of) the "Spoken Wikipedia Corpora" for train.py
Use "python3 import_swc.py -h" for help
"""
@ -22,8 +22,8 @@ from multiprocessing.pool import ThreadPool
import progressbar
import sox
from deepspeech_training.util.downloader import SIMPLE_BAR, maybe_download
from deepspeech_training.util.importers import validate_label_eng as validate_label
from coqui_stt_training.util.downloader import SIMPLE_BAR, maybe_download
from coqui_stt_training.util.importers import validate_label_eng as validate_label
from ds_ctcdecoder import Alphabet
SWC_URL = "https://www2.informatik.uni-hamburg.de/nats/pub/SWC/SWC_{language}.tar"

4
bin/import_ted.py
View File

@ -10,8 +10,8 @@ import pandas
from sox import Transformer
from tensorflow.python.platform import gfile
from deepspeech_training.util.downloader import maybe_download
from deepspeech_training.util.stm import parse_stm_file
from coqui_stt_training.util.downloader import maybe_download
from coqui_stt_training.util.stm import parse_stm_file
def _download_and_preprocess_data(data_dir):

8
bin/import_ts.py
View File

@ -10,8 +10,8 @@ import progressbar
import sox
import unidecode
from deepspeech_training.util.downloader import SIMPLE_BAR, maybe_download
from deepspeech_training.util.importers import (
from coqui_stt_training.util.downloader import SIMPLE_BAR, maybe_download
from coqui_stt_training.util.importers import (
get_counter,
get_imported_samples,
get_importers_parser,
@ -25,7 +25,7 @@ MAX_SECS = 15
ARCHIVE_NAME = "2019-04-11_fr_FR"
ARCHIVE_DIR_NAME = "ts_" + ARCHIVE_NAME
ARCHIVE_URL = (
"https://deepspeech-storage-mirror.s3.fr-par.scw.cloud/" + ARCHIVE_NAME + ".zip"
"https://Coqui STT-storage-mirror.s3.fr-par.scw.cloud/" + ARCHIVE_NAME + ".zip"
)
@ -38,7 +38,7 @@ def _download_and_preprocess_data(target_dir, english_compatible=False):
)
# Conditionally extract archive data
_maybe_extract(target_dir, ARCHIVE_DIR_NAME, archive_path)
# Conditionally convert TrainingSpeech data to DeepSpeech CSVs and wav
# Conditionally convert TrainingSpeech data to Coqui STT CSVs and wav
_maybe_convert_sets(
target_dir, ARCHIVE_DIR_NAME, english_compatible=english_compatible
)

6
bin/import_tuda.py
View File

@ -1,6 +1,6 @@
#!/usr/bin/env python
"""
Downloads and prepares (parts of) the "German Distant Speech" corpus (TUDA) for DeepSpeech.py
Downloads and prepares (parts of) the "German Distant Speech" corpus (TUDA) for train.py
Use "python3 import_tuda.py -h" for help
"""
import argparse
@ -14,8 +14,8 @@ from collections import Counter
import progressbar
from deepspeech_training.util.downloader import SIMPLE_BAR, maybe_download
from deepspeech_training.util.importers import validate_label_eng as validate_label
from coqui_stt_training.util.downloader import SIMPLE_BAR, maybe_download
from coqui_stt_training.util.importers import validate_label_eng as validate_label
from ds_ctcdecoder import Alphabet
TUDA_VERSION = "v2"

6
bin/import_vctk.py
View File

@ -11,8 +11,8 @@ from zipfile import ZipFile
import librosa
import progressbar
from deepspeech_training.util.downloader import SIMPLE_BAR, maybe_download
from deepspeech_training.util.importers import (
from coqui_stt_training.util.downloader import SIMPLE_BAR, maybe_download
from coqui_stt_training.util.importers import (
get_counter,
get_imported_samples,
print_import_report,
@ -35,7 +35,7 @@ def _download_and_preprocess_data(target_dir):
archive_path = maybe_download(ARCHIVE_NAME, target_dir, ARCHIVE_URL)
# Conditionally extract common voice data
_maybe_extract(target_dir, ARCHIVE_DIR_NAME, archive_path)
# Conditionally convert common voice CSV files and mp3 data to DeepSpeech CSVs and wav
# Conditionally convert common voice CSV files and mp3 data to Coqui STT CSVs and wav
_maybe_convert_sets(target_dir, ARCHIVE_DIR_NAME)

2
bin/import_voxforge.py
View File

@ -14,7 +14,7 @@ from os import makedirs, path
import pandas
from bs4 import BeautifulSoup
from tensorflow.python.platform import gfile
from deepspeech_training.util.downloader import maybe_download
from coqui_stt_training.util.downloader import maybe_download
"""The number of jobs to run in parallel"""
NUM_PARALLEL = 8

10
bin/play.py
View File

@ -1,6 +1,6 @@
#!/usr/bin/env python
"""
Tool for playing (and augmenting) single samples or samples from Sample Databases (SDB files) and DeepSpeech CSV files
Tool for playing (and augmenting) single samples or samples from Sample Databases (SDB files) and 🐸STT CSV files
Use "python3 play.py -h" for help
"""
@ -9,9 +9,9 @@ import sys
import random
import argparse
from deepspeech_training.util.audio import get_loadable_audio_type_from_extension, AUDIO_TYPE_PCM, AUDIO_TYPE_WAV
from deepspeech_training.util.sample_collections import SampleList, LabeledSample, samples_from_source
from deepspeech_training.util.augmentations import parse_augmentations, apply_sample_augmentations, SampleAugmentation
from coqui_stt_training.util.audio import get_loadable_audio_type_from_extension, AUDIO_TYPE_PCM, AUDIO_TYPE_WAV
from coqui_stt_training.util.sample_collections import SampleList, LabeledSample, samples_from_source
from coqui_stt_training.util.augmentations import parse_augmentations, apply_sample_augmentations, SampleAugmentation
def get_samples_in_play_order():
@ -68,7 +68,7 @@ def play_collection():
def handle_args():
parser = argparse.ArgumentParser(
description="Tool for playing (and augmenting) single samples or samples from Sample Databases (SDB files) "
"and DeepSpeech CSV files"
"and Coqui STT CSV files"
)
parser.add_argument("source", help="Sample DB, CSV or WAV file to play samples from")
parser.add_argument(

8
bin/run-ldc93s1.sh
View File

@ -1,7 +1,7 @@
#!/bin/sh
set -xe
if [ ! -f DeepSpeech.py ]; then
echo "Please make sure you run this from DeepSpeech's top level directory."
if [ ! -f train.py ]; then
echo "Please make sure you run this from STT's top level directory."
exit 1
fi;
@ -13,14 +13,14 @@ fi;
if [ -d "${COMPUTE_KEEP_DIR}" ]; then
checkpoint_dir=$COMPUTE_KEEP_DIR
else
checkpoint_dir=$(python -c 'from xdg import BaseDirectory as xdg; print(xdg.save_data_path("deepspeech/ldc93s1"))')
checkpoint_dir=$(python -c 'from xdg import BaseDirectory as xdg; print(xdg.save_data_path("stt/ldc93s1"))')
fi
# Force only one visible device because we have a single-sample dataset
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar \
python -u train.py --noshow_progressbar \
--train_files data/ldc93s1/ldc93s1.csv \
--test_files data/ldc93s1/ldc93s1.csv \
--train_batch_size 1 \

2
bin/run-tc-graph_augmentations.sh
View File

@ -14,7 +14,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--train_files ${ldc93s1_csv} --train_batch_size 1 \
--scorer "" \
--augment dropout \

2
bin/run-tc-ldc93s1_checkpoint.sh
View File

@ -14,7 +14,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--train_files ${ldc93s1_csv} --train_batch_size 1 \
--dev_files ${ldc93s1_csv} --dev_batch_size 1 \
--test_files ${ldc93s1_csv} --test_batch_size 1 \

2
bin/run-tc-ldc93s1_checkpoint_bytes.sh
View File

@ -14,7 +14,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--train_files ${ldc93s1_csv} --train_batch_size 1 \
--dev_files ${ldc93s1_csv} --dev_batch_size 1 \
--test_files ${ldc93s1_csv} --test_batch_size 1 \

2
bin/run-tc-ldc93s1_checkpoint_sdb.sh
View File

@ -20,7 +20,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--train_files ${ldc93s1_sdb} --train_batch_size 1 \
--dev_files ${ldc93s1_sdb} --dev_batch_size 1 \
--test_files ${ldc93s1_sdb} --test_batch_size 1 \

2
bin/run-tc-ldc93s1_new.sh
View File

@ -17,7 +17,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--train_files ${ldc93s1_csv} --train_batch_size 1 \
--feature_cache '/tmp/ldc93s1_cache' \
--dev_files ${ldc93s1_csv} --dev_batch_size 1 \

2
bin/run-tc-ldc93s1_new_bytes.sh
View File

@ -17,7 +17,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--train_files ${ldc93s1_csv} --train_batch_size 1 \
--feature_cache '/tmp/ldc93s1_cache' \
--dev_files ${ldc93s1_csv} --dev_batch_size 1 \

2
bin/run-tc-ldc93s1_new_bytes_tflite.sh
View File

@ -16,7 +16,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar \
python -u train.py --noshow_progressbar \
--n_hidden 100 \
--checkpoint_dir '/tmp/ckpt_bytes' \
--export_dir '/tmp/train_bytes_tflite' \

2
bin/run-tc-ldc93s1_new_metrics.sh
View File

@ -17,7 +17,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--train_files ${ldc93s1_csv} --train_batch_size 1 \
--dev_files ${ldc93s1_csv} --dev_batch_size 1 \
--test_files ${ldc93s1_csv} --test_batch_size 1 \

2
bin/run-tc-ldc93s1_new_sdb.sh
View File

@ -23,7 +23,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--train_files ${ldc93s1_sdb} --train_batch_size 1 \
--dev_files ${ldc93s1_sdb} --dev_batch_size 1 \
--test_files ${ldc93s1_sdb} --test_batch_size 1 \

2
bin/run-tc-ldc93s1_new_sdb_csv.sh
View File

@ -23,7 +23,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--train_files ${ldc93s1_sdb},${ldc93s1_csv} --train_batch_size 1 \
--feature_cache '/tmp/ldc93s1_cache_sdb_csv' \
--dev_files ${ldc93s1_sdb},${ldc93s1_csv} --dev_batch_size 1 \

4
bin/run-tc-ldc93s1_singleshotinference.sh
View File

@ -14,7 +14,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--train_files ${ldc93s1_csv} --train_batch_size 1 \
--dev_files ${ldc93s1_csv} --dev_batch_size 1 \
--test_files ${ldc93s1_csv} --test_batch_size 1 \
@ -23,7 +23,7 @@ python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
--learning_rate 0.001 --dropout_rate 0.05 \
--scorer_path 'data/smoke_test/pruned_lm.scorer'
python -u DeepSpeech.py \
python -u train.py \
--n_hidden 100 \
--checkpoint_dir '/tmp/ckpt' \
--scorer_path 'data/smoke_test/pruned_lm.scorer' \

4
bin/run-tc-ldc93s1_tflite.sh
View File

@ -16,7 +16,7 @@ fi;
# and when trying to run on multiple devices (like GPUs), this will break
export CUDA_VISIBLE_DEVICES=0
python -u DeepSpeech.py --noshow_progressbar \
python -u train.py --noshow_progressbar \
--n_hidden 100 \
--checkpoint_dir '/tmp/ckpt' \
--export_dir '/tmp/train_tflite' \
@ -26,7 +26,7 @@ python -u DeepSpeech.py --noshow_progressbar \
mkdir /tmp/train_tflite/en-us
python -u DeepSpeech.py --noshow_progressbar \
python -u train.py --noshow_progressbar \
--n_hidden 100 \
--checkpoint_dir '/tmp/ckpt' \
--export_dir '/tmp/train_tflite/en-us' \

6
bin/run-tc-transfer.sh
View File

@ -29,7 +29,7 @@ for LOAD in 'init' 'last' 'auto'; do
echo "########################################################"
echo "#### Train ENGLISH model with just --checkpoint_dir ####"
echo "########################################################"
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--alphabet_config_path "./data/alphabet.txt" \
--load_train "$LOAD" \
--train_files "${ldc93s1_csv}" --train_batch_size 1 \
@ -43,7 +43,7 @@ for LOAD in 'init' 'last' 'auto'; do
echo "##############################################################################"
echo "#### Train ENGLISH model with --save_checkpoint_dir --load_checkpoint_dir ####"
echo "##############################################################################"
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--alphabet_config_path "./data/alphabet.txt" \
--load_train "$LOAD" \
--train_files "${ldc93s1_csv}" --train_batch_size 1 \
@ -58,7 +58,7 @@ for LOAD in 'init' 'last' 'auto'; do
echo "####################################################################################"
echo "#### Transfer to RUSSIAN model with --save_checkpoint_dir --load_checkpoint_dir ####"
echo "####################################################################################"
python -u DeepSpeech.py --noshow_progressbar --noearly_stop \
python -u train.py --noshow_progressbar --noearly_stop \
--drop_source_layers 1 \
--alphabet_config_path "${ru_dir}/alphabet.ru" \
--load_train 'last' \

4
data/README.rst
View File

@ -3,9 +3,9 @@ Language-Specific Data
This directory contains language-specific data files. Most importantly, you will find here:
1. A list of unique characters for the target language (e.g. English) in ``data/alphabet.txt``. After installing the training code, you can check ``python -m deepspeech_training.util.check_characters --help`` for a tool that creates an alphabet file from a list of training CSV files.
1. A list of unique characters for the target language (e.g. English) in ``data/alphabet.txt``. After installing the training code, you can check ``python -m coqui_stt_training.util.check_characters --help`` for a tool that creates an alphabet file from a list of training CSV files.
2. A script used to generate a binary n-gram language model: ``data/lm/generate_lm.py``.
For more information on how to build these resources from scratch, see the ``External scorer scripts`` section on `deepspeech.readthedocs.io <https://deepspeech.readthedocs.io/>`_.
For more information on how to build these resources from scratch, see the ``External scorer scripts`` section on `stt.readthedocs.io <https://stt.readthedocs.io/>`_.

2
data/lm/generate_lm.py
View File

@ -130,7 +130,7 @@ def build_lm(args, data_lower, vocab_str):
def main():
parser = argparse.ArgumentParser(
description="Generate lm.binary and top-k vocab for DeepSpeech."
description="Generate lm.binary and top-k vocab for Coqui STT."
)
parser.add_argument(
"--input_txt",

4
doc/DeepSpeech.rst → doc/Architecture.rst
View File

@ -1,5 +1,5 @@
DeepSpeech Model
================
STT Model
=========
The aim of this project is to create a simple, open, and ubiquitous speech
recognition engine. Simple, in that the engine should not require server-class

128
doc/BUILDING.rst
View File

@ -1,12 +1,12 @@
.. _build-native-client:
Building DeepSpeech Binaries
============================
Building Coqui STT Binaries
===========================
This section describes how to rebuild binaries. We have already several prebuilt binaries for all the supported platform,
it is highly advised to use them except if you know what you are doing.
If you'd like to build the DeepSpeech binaries yourself, you'll need the following pre-requisites downloaded and installed:
If you'd like to build the 🐸STT binaries yourself, you'll need the following pre-requisites downloaded and installed:
* `Bazel 3.1.0 <https://github.com/bazelbuild/bazel/releases/tag/3.1.0>`_
* `General TensorFlow r2.3 requirements <https://www.tensorflow.org/install/source#tested_build_configurations>`_
@ -26,18 +26,18 @@ If you'd like to build the language bindings or the decoder package, you'll also
Dependencies
------------
If you follow these instructions, you should compile your own binaries of DeepSpeech (built on TensorFlow using Bazel).
If you follow these instructions, you should compile your own binaries of 🐸STT (built on TensorFlow using Bazel).
For more information on configuring TensorFlow, read the docs up to the end of `"Configure the Build" <https://www.tensorflow.org/install/source#configure_the_build>`_.
Checkout source code
^^^^^^^^^^^^^^^^^^^^
Clone DeepSpeech source code (TensorFlow will come as a submdule):
Clone 🐸STT source code (TensorFlow will come as a submdule):
.. code-block::
git clone https://github.com/mozilla/DeepSpeech.git
git clone https://github.com/coqui-ai/STT.git
git submodule sync tensorflow/
git submodule update --init tensorflow/
@ -56,24 +56,24 @@ After you have installed the correct version of Bazel, configure TensorFlow:
cd tensorflow
./configure
Compile DeepSpeech
------------------
Compile Coqui STT
-----------------
Compile ``libdeepspeech.so``
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Compile ``libstt.so``
^^^^^^^^^^^^^^^^^^^^^
Within your TensorFlow directory, there should be a symbolic link to the DeepSpeech ``native_client`` directory. If it is not present, create it with the follow command:
Within your TensorFlow directory, there should be a symbolic link to the 🐸STT ``native_client`` directory. If it is not present, create it with the follow command:
.. code-block::
cd tensorflow
ln -s ../native_client
You can now use Bazel to build the main DeepSpeech library, ``libdeepspeech.so``. Add ``--config=cuda`` if you want a CUDA build.
You can now use Bazel to build the main 🐸STT library, ``libstt.so``. Add ``--config=cuda`` if you want a CUDA build.
.. code-block::
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic -c opt --copt=-O3 --copt="-D_GLIBCXX_USE_CXX11_ABI=0" --copt=-fvisibility=hidden //native_client:libdeepspeech.so
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic -c opt --copt=-O3 --copt="-D_GLIBCXX_USE_CXX11_ABI=0" --copt=-fvisibility=hidden //native_client:libstt.so
The generated binaries will be saved to ``bazel-bin/native_client/``.
@ -82,24 +82,24 @@ The generated binaries will be saved to ``bazel-bin/native_client/``.
Compile ``generate_scorer_package``
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Following the same setup as for ``libdeepspeech.so`` above, you can rebuild the ``generate_scorer_package`` binary by adding its target to the command line: ``//native_client:generate_scorer_package``.
Following the same setup as for ``libstt.so`` above, you can rebuild the ``generate_scorer_package`` binary by adding its target to the command line: ``//native_client:generate_scorer_package``.
Using the example from above you can build the library and that binary at the same time:
.. code-block::
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic -c opt --copt=-O3 --copt="-D_GLIBCXX_USE_CXX11_ABI=0" --copt=-fvisibility=hidden //native_client:libdeepspeech.so //native_client:generate_scorer_package
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic -c opt --copt=-O3 --copt="-D_GLIBCXX_USE_CXX11_ABI=0" --copt=-fvisibility=hidden //native_client:libstt.so //native_client:generate_scorer_package
The generated binaries will be saved to ``bazel-bin/native_client/``.
Compile Language Bindings
^^^^^^^^^^^^^^^^^^^^^^^^^
Now, ``cd`` into the ``DeepSpeech/native_client`` directory and use the ``Makefile`` to build all the language bindings (C++ client, Python package, Nodejs package, etc.).
Now, ``cd`` into the ``STT/native_client`` directory and use the ``Makefile`` to build all the language bindings (C++ client, Python package, Nodejs package, etc.).
.. code-block::
cd ../DeepSpeech/native_client
make deepspeech
cd ../STT/native_client
make stt
Installing your own Binaries
----------------------------
@ -121,9 +121,9 @@ Included are a set of generated Python bindings. After following the above build
cd native_client/python
make bindings
pip install dist/deepspeech*
pip install dist/stt-*
The API mirrors the C++ API and is demonstrated in `client.py <python/client.py>`_. Refer to `deepspeech.h <deepspeech.h>`_ for documentation.
The API mirrors the C++ API and is demonstrated in `client.py <python/client.py>`_. Refer to `coqui-stt.h <coqui-stt.h>`_ for documentation.
Install NodeJS / ElectronJS bindings
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@ -136,7 +136,7 @@ After following the above build and installation instructions, the Node.JS bindi
make build
make npm-pack
This will create the package ``deepspeech-VERSION.tgz`` in ``native_client/javascript``.
This will create the package ``stt-VERSION.tgz`` in ``native_client/javascript``.
Install the CTC decoder package
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@ -196,23 +196,23 @@ So your command line for ``RPi3`` and ``ARMv7`` should look like:
.. code-block::
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic --config=rpi3 --config=rpi3_opt -c opt --copt=-O3 --copt=-fvisibility=hidden //native_client:libdeepspeech.so
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic --config=rpi3 --config=rpi3_opt -c opt --copt=-O3 --copt=-fvisibility=hidden //native_client:libstt.so
And your command line for ``LePotato`` and ``ARM64`` should look like:
.. code-block::
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic --config=rpi3-armv8 --config=rpi3-armv8_opt -c opt --copt=-O3 --copt=-fvisibility=hidden //native_client:libdeepspeech.so
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic --config=rpi3-armv8 --config=rpi3-armv8_opt -c opt --copt=-O3 --copt=-fvisibility=hidden //native_client:libstt.so
While we test only on RPi3 Raspbian Buster and LePotato ARMBian Buster, anything compatible with ``armv7-a cortex-a53`` or ``armv8-a cortex-a53`` should be fine.
The ``deepspeech`` binary can also be cross-built, with ``TARGET=rpi3`` or ``TARGET=rpi3-armv8``. This might require you to setup a system tree using the tool ``multistrap`` and the multitrap configuration files: ``native_client/multistrap_armbian64_buster.conf`` and ``native_client/multistrap_raspbian_buster.conf``.
The ``stt`` binary can also be cross-built, with ``TARGET=rpi3`` or ``TARGET=rpi3-armv8``. This might require you to setup a system tree using the tool ``multistrap`` and the multitrap configuration files: ``native_client/multistrap_armbian64_buster.conf`` and ``native_client/multistrap_raspbian_buster.conf``.
The path of the system tree can be overridden from the default values defined in ``definitions.mk`` through the ``RASPBIAN`` ``make`` variable.
.. code-block::
cd ../DeepSpeech/native_client
make TARGET=<system> deepspeech
cd ../STT/native_client
make TARGET=<system> stt
Android devices support
-----------------------
@ -224,64 +224,66 @@ Please refer to TensorFlow documentation on how to setup the environment to buil
Using the library from Android project
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
We provide uptodate and tested ``libdeepspeech`` usable as an ``AAR`` package,
for Android versions starting with 7.0 to 11.0. The package is published on
`JCenter <https://bintray.com/alissy/org.mozilla.deepspeech/libdeepspeech>`_,
and the ``JCenter`` repository should be available by default in any Android
project. Please make sure your project is setup to pull from this repository.
You can then include the library by just adding this line to your
``gradle.build``, adjusting ``VERSION`` to the version you need:
Due to the discontinuation of Bintray JCenter we do not have pre-built Android packages published for now. We are working to move to Maven Central and will update this section when it's available.
.. We provide uptodate and tested ``libstt`` usable as an ``AAR`` package,
for Android versions starting with 7.0 to 11.0. The package is published on
`JCenter <https://bintray.com/coqui/ai.coqui.stt/libstt>`_,
and the ``JCenter`` repository should be available by default in any Android
project. Please make sure your project is setup to pull from this repository.
You can then include the library by just adding this line to your
``gradle.build``, adjusting ``VERSION`` to the version you need:
.. code-block::
implementation 'stt.coqui.ai:libstt:VERSION@aar'
Building ``libstt.so`` for Android
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
You can build the ``libstt.so`` using (ARMv7):
.. code-block::
implementation 'deepspeech.mozilla.org:libdeepspeech:VERSION@aar'
Building ``libdeepspeech.so``
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
You can build the ``libdeepspeech.so`` using (ARMv7):
.. code-block::
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic --config=android --config=android_arm --define=runtime=tflite --action_env ANDROID_NDK_API_LEVEL=21 --cxxopt=-std=c++14 --copt=-D_GLIBCXX_USE_C99 //native_client:libdeepspeech.so
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic --config=android --config=android_arm --define=runtime=tflite --action_env ANDROID_NDK_API_LEVEL=21 --cxxopt=-std=c++14 --copt=-D_GLIBCXX_USE_C99 //native_client:libstt.so
Or (ARM64):
.. code-block::
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic --config=android --config=android_arm64 --define=runtime=tflite --action_env ANDROID_NDK_API_LEVEL=21 --cxxopt=-std=c++14 --copt=-D_GLIBCXX_USE_C99 //native_client:libdeepspeech.so
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" --config=monolithic --config=android --config=android_arm64 --define=runtime=tflite --action_env ANDROID_NDK_API_LEVEL=21 --cxxopt=-std=c++14 --copt=-D_GLIBCXX_USE_C99 //native_client:libstt.so
Building ``libdeepspeech.aar``
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Building ``libstt.aar``
^^^^^^^^^^^^^^^^^^^^^^^
In the unlikely event you have to rebuild the JNI bindings, source code is
available under the ``libdeepspeech`` subdirectory. Building depends on shared
object: please ensure to place ``libdeepspeech.so`` into the
``libdeepspeech/libs/{arm64-v8a,armeabi-v7a,x86_64}/`` matching subdirectories.
available under the ``libstt`` subdirectory. Building depends on shared
object: please ensure to place ``libstt.so`` into the
``libstt/libs/{arm64-v8a,armeabi-v7a,x86_64}/`` matching subdirectories.
Building the bindings is managed by ``gradle`` and should be limited to issuing
``./gradlew libdeepspeech:build``, producing an ``AAR`` package in
``./libdeepspeech/build/outputs/aar/``.
``./gradlew libstt:build``, producing an ``AAR`` package in
``./libstt/build/outputs/aar/``.
Please note that you might have to copy the file to a local Maven repository
and adapt file naming (when missing, the error message should states what
filename it expects and where).
Building C++ ``deepspeech`` binary
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Building C++ ``stt`` binary
^^^^^^^^^^^^^^^^^^^^^^^^^^^
Building the ``deepspeech`` binary will happen through ``ndk-build`` (ARMv7):
Building the ``stt`` binary will happen through ``ndk-build`` (ARMv7):
.. code-block::
cd ../DeepSpeech/native_client
cd ../STT/native_client
$ANDROID_NDK_HOME/ndk-build APP_PLATFORM=android-21 APP_BUILD_SCRIPT=$(pwd)/Android.mk NDK_PROJECT_PATH=$(pwd) APP_STL=c++_shared TFDIR=$(pwd)/../tensorflow/ TARGET_ARCH_ABI=armeabi-v7a
And (ARM64):
.. code-block::
cd ../DeepSpeech/native_client
cd ../STT/native_client
$ANDROID_NDK_HOME/ndk-build APP_PLATFORM=android-21 APP_BUILD_SCRIPT=$(pwd)/Android.mk NDK_PROJECT_PATH=$(pwd) APP_STL=c++_shared TFDIR=$(pwd)/../tensorflow/ TARGET_ARCH_ABI=arm64-v8a
Android demo APK
@ -303,13 +305,13 @@ demo of one usage of the application. For example, it's only able to read PCM
mono 16kHz 16-bits file and it might fail on some WAVE file that are not
following exactly the specification.
Running ``deepspeech`` via adb
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Running ``stt`` via adb
^^^^^^^^^^^^^^^^^^^^^^^
You should use ``adb push`` to send data to device, please refer to Android
documentation on how to use that.
Please push DeepSpeech data to ``/sdcard/deepspeech/``\ , including:
Please push 🐸STT data to ``/sdcard/STT/``\ , including:
* ``output_graph.tflite`` which is the TF Lite model
@ -319,8 +321,8 @@ Please push DeepSpeech data to ``/sdcard/deepspeech/``\ , including:
Then, push binaries from ``native_client.tar.xz`` to ``/data/local/tmp/ds``\ :
* ``deepspeech``
* ``libdeepspeech.so``
* ``stt``
* ``libstt.so``
* ``libc++_shared.so``
You should then be able to run as usual, using a shell from ``adb shell``\ :
@ -328,7 +330,7 @@ You should then be able to run as usual, using a shell from ``adb shell``\ :
.. code-block::
user@device$ cd /data/local/tmp/ds/
user@device$ LD_LIBRARY_PATH=$(pwd)/ ./deepspeech [...]
user@device$ LD_LIBRARY_PATH=$(pwd)/ ./stt [...]
Please note that Android linker does not support ``rpath`` so you have to set
``LD_LIBRARY_PATH``. Properly wrapped / packaged bindings does embed the library
@ -347,7 +349,7 @@ to leverage GPU / DSP / NPU * Hexagon, the Qualcomm-specific DSP
This is highly experimental:
* Requires passing environment variable ``DS_TFLITE_DELEGATE`` with values of
* Requires passing environment variable ``STT_TFLITE_DELEGATE`` with values of
``gpu``, ``nnapi`` or ``hexagon`` (only one at a time)
* Might require exported model changes (some Op might not be supported)
* We can't guarantee it will work, nor it will be faster than default

26
doc/BUILDING_DotNet.rst
View File

@ -1,9 +1,9 @@
.. _build-native-client-dotnet:
Building DeepSpeech native client for Windows
=============================================
Building Coqui STT native client for Windows
============================================
Now we can build the native client of DeepSpeech and run inference on Windows using the C# client, to do that we need to compile the ``native_client``.
Now we can build the native client of 🐸STT and run inference on Windows using the C# client, to do that we need to compile the ``native_client``.
**Table of Contents**
@ -44,11 +44,11 @@ We highly recommend sticking to the recommended versions of CUDA/cuDNN in order
Getting the code
----------------
We need to clone ``mozilla/DeepSpeech``.
We need to clone ``coqui-ai/STT``.
.. code-block:: bash
git clone https://github.com/mozilla/DeepSpeech
git clone https://github.com/coqui-ai/STT
git submodule sync tensorflow/
git submodule update --init tensorflow/
@ -61,8 +61,8 @@ There should already be a symbolic link, for this example let's suppose that we
.
├── D:\
│ ├── cloned # Contains DeepSpeech and tensorflow side by side
│ │ └── DeepSpeech # Root of the cloned DeepSpeech
│ ├── cloned # Contains 🐸STT and tensorflow side by side
│ │ └── STT # Root of the cloned 🐸STT
│ │ ├── tensorflow # Root of the cloned mozilla/tensorflow
└── ...
@ -71,7 +71,7 @@ Change your path accordingly to your path structure, for the structure above we
.. code-block:: bash
mklink /d "D:\cloned\DeepSpeech\tensorflow\native_client" "D:\cloned\DeepSpeech\native_client"
mklink /d "D:\cloned\STT\tensorflow\native_client" "D:\cloned\STT\native_client"
Adding environment variables
----------------------------
@ -119,7 +119,7 @@ Building the native_client
There's one last command to run before building, you need to run the `configure.py <https://github.com/mozilla/tensorflow/blob/master/configure.py>`_ inside ``tensorflow`` cloned directory.
At this point we are ready to start building the ``native_client``, go to ``tensorflow`` sub-directory, following our examples should be ``D:\cloned\DeepSpeech\tensorflow``.
At this point we are ready to start building the ``native_client``, go to ``tensorflow`` sub-directory, following our examples should be ``D:\cloned\STT\tensorflow``.
CPU
~~~
@ -128,7 +128,7 @@ We will add AVX/AVX2 support in the command, please make sure that your CPU supp
.. code-block:: bash
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" -c opt --copt=/arch:AVX --copt=/arch:AVX2 //native_client:libdeepspeech.so
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" -c opt --copt=/arch:AVX --copt=/arch:AVX2 //native_client:libstt.so
GPU with CUDA
~~~~~~~~~~~~~
@ -137,11 +137,11 @@ If you enabled CUDA in `configure.py <https://github.com/mozilla/tensorflow/blob
.. code-block:: bash
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" -c opt --config=cuda --copt=/arch:AVX --copt=/arch:AVX2 //native_client:libdeepspeech.so
bazel build --workspace_status_command="bash native_client/bazel_workspace_status_cmd.sh" -c opt --config=cuda --copt=/arch:AVX --copt=/arch:AVX2 //native_client:libstt.so
Be patient, if you enabled AVX/AVX2 and CUDA it will take a long time. Finally you should see it stops and shows the path to the generated ``libdeepspeech.so``.
Be patient, if you enabled AVX/AVX2 and CUDA it will take a long time. Finally you should see it stops and shows the path to the generated ``libstt.so``.
Using the generated library
---------------------------
As for now we can only use the generated ``libdeepspeech.so`` with the C# clients, go to `native_client/dotnet/ <https://github.com/mozilla/DeepSpeech/tree/master/native_client/dotnet>`_ in your DeepSpeech directory and open the Visual Studio solution, then we need to build in debug or release mode, finally we just need to copy ``libdeepspeech.so`` to the generated ``x64/Debug`` or ``x64/Release`` directory.
As for now we can only use the generated ``libstt.so`` with the C# clients, go to `native_client/dotnet/ <https://github.com/coqui-ai/STT/tree/main/native_client/dotnet>`_ in your STT directory and open the Visual Studio solution, then we need to build in debug or release mode, finally we just need to copy ``libstt.so`` to the generated ``x64/Debug`` or ``x64/Release`` directory.

84
doc/C-API.rst
View File

@ -10,65 +10,65 @@ C API
See also the list of error codes including descriptions for each error in :ref:`error-codes`.
.. doxygenfunction:: DS_CreateModel
:project: deepspeech-c
.. doxygenfunction:: STT_CreateModel
:project: stt-c
.. doxygenfunction:: DS_FreeModel
:project: deepspeech-c
.. doxygenfunction:: STT_FreeModel
:project: stt-c
.. doxygenfunction:: DS_EnableExternalScorer
:project: deepspeech-c
.. doxygenfunction:: STT_EnableExternalScorer
:project: stt-c
.. doxygenfunction:: DS_DisableExternalScorer
:project: deepspeech-c
.. doxygenfunction:: STT_DisableExternalScorer
:project: stt-c
.. doxygenfunction:: DS_AddHotWord
:project: deepspeech-c
.. doxygenfunction:: STT_AddHotWord
:project: stt-c
.. doxygenfunction:: DS_EraseHotWord
:project: deepspeech-c
.. doxygenfunction:: STT_EraseHotWord
:project: stt-c
.. doxygenfunction:: DS_ClearHotWords
:project: deepspeech-c
.. doxygenfunction:: STT_ClearHotWords
:project: stt-c
.. doxygenfunction:: DS_SetScorerAlphaBeta
:project: deepspeech-c
.. doxygenfunction:: STT_SetScorerAlphaBeta
:project: stt-c
.. doxygenfunction:: DS_GetModelSampleRate
:project: deepspeech-c
.. doxygenfunction:: STT_GetModelSampleRate
:project: stt-c
.. doxygenfunction:: DS_SpeechToText
:project: deepspeech-c
.. doxygenfunction:: STT_SpeechToText
:project: stt-c
.. doxygenfunction:: DS_SpeechToTextWithMetadata
:project: deepspeech-c
.. doxygenfunction:: STT_SpeechToTextWithMetadata
:project: stt-c
.. doxygenfunction:: DS_CreateStream
:project: deepspeech-c
.. doxygenfunction:: STT_CreateStream
:project: stt-c
.. doxygenfunction:: DS_FeedAudioContent
:project: deepspeech-c
.. doxygenfunction:: STT_FeedAudioContent
:project: stt-c
.. doxygenfunction:: DS_IntermediateDecode
:project: deepspeech-c
.. doxygenfunction:: STT_IntermediateDecode
:project: stt-c
.. doxygenfunction:: DS_IntermediateDecodeWithMetadata
:project: deepspeech-c
.. doxygenfunction:: STT_IntermediateDecodeWithMetadata
:project: stt-c
.. doxygenfunction:: DS_FinishStream
:project: deepspeech-c
.. doxygenfunction:: STT_FinishStream
:project: stt-c
.. doxygenfunction:: DS_FinishStreamWithMetadata
:project: deepspeech-c
.. doxygenfunction:: STT_FinishStreamWithMetadata
:project: stt-c
.. doxygenfunction:: DS_FreeStream
:project: deepspeech-c
.. doxygenfunction:: STT_FreeStream
:project: stt-c
.. doxygenfunction:: DS_FreeMetadata
:project: deepspeech-c
.. doxygenfunction:: STT_FreeMetadata
:project: stt-c
.. doxygenfunction:: DS_FreeString
:project: deepspeech-c
.. doxygenfunction:: STT_FreeString
:project: stt-c
.. doxygenfunction:: DS_Version
:project: deepspeech-c
.. doxygenfunction:: STT_Version
:project: stt-c

2
doc/Contributed-Examples.rst
View File

@ -1,4 +1,4 @@
User contributed examples
=========================
There are also several user contributed examples available on a separate examples repository: `https://github.com/mozilla/DeepSpeech-examples <https://github.com/mozilla/DeepSpeech-examples>`_.
There are also several user contributed examples available on a separate examples repository: `https://github.com/coqui-ai/STT-examples <https://github.com/coqui-ai/STT-examples>`_.

12
doc/Decoder.rst
View File

@ -6,7 +6,7 @@ CTC beam search decoder
Introduction
^^^^^^^^^^^^
DeepSpeech uses the `Connectionist Temporal Classification <http://www.cs.toronto.edu/~graves/icml_2006.pdf>`_ loss function. For an excellent explanation of CTC and its usage, see this Distill article: `Sequence Modeling with CTC <https://distill.pub/2017/ctc/>`_. This document assumes the reader is familiar with the concepts described in that article, and describes DeepSpeech specific behaviors that developers building systems with DeepSpeech should know to avoid problems.
🐸STT uses the `Connectionist Temporal Classification <http://www.cs.toronto.edu/~graves/icml_2006.pdf>`_ loss function. For an excellent explanation of CTC and its usage, see this Distill article: `Sequence Modeling with CTC <https://distill.pub/2017/ctc/>`_. This document assumes the reader is familiar with the concepts described in that article, and describes 🐸STT specific behaviors that developers building systems with 🐸STT should know to avoid problems.
Note: Documentation for the tooling for creating custom scorer packages is available in :ref:`scorer-scripts`.
@ -16,19 +16,19 @@ The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "S
External scorer
^^^^^^^^^^^^^^^
DeepSpeech clients support OPTIONAL use of an external language model to improve the accuracy of the predicted transcripts. In the code, command line parameters, and documentation, this is referred to as a "scorer". The scorer is used to compute the likelihood (also called a score, hence the name "scorer") of sequences of words or characters in the output, to guide the decoder towards more likely results. This improves accuracy significantly.
🐸STT clients support OPTIONAL use of an external language model to improve the accuracy of the predicted transcripts. In the code, command line parameters, and documentation, this is referred to as a "scorer". The scorer is used to compute the likelihood (also called a score, hence the name "scorer") of sequences of words or characters in the output, to guide the decoder towards more likely results. This improves accuracy significantly.
The use of an external scorer is fully optional. When an external scorer is not specified, DeepSpeech still uses a beam search decoding algorithm, but without any outside scoring.
The use of an external scorer is fully optional. When an external scorer is not specified, 🐸STT still uses a beam search decoding algorithm, but without any outside scoring.
Currently, the DeepSpeech external scorer is implemented with `KenLM <https://kheafield.com/code/kenlm/>`_, plus some tooling to package the necessary files and metadata into a single ``.scorer`` package. The tooling lives in ``data/lm/``. The scripts included in ``data/lm/`` can be used and modified to build your own language model based on your particular use case or language. See :ref:`scorer-scripts` for more details on how to reproduce our scorer file as well as create your own.
Currently, the 🐸STT external scorer is implemented with `KenLM <https://kheafield.com/code/kenlm/>`_, plus some tooling to package the necessary files and metadata into a single ``.scorer`` package. The tooling lives in ``data/lm/``. The scripts included in ``data/lm/`` can be used and modified to build your own language model based on your particular use case or language. See :ref:`scorer-scripts` for more details on how to reproduce our scorer file as well as create your own.
The scripts are geared towards replicating the language model files we release as part of `DeepSpeech model releases <https://github.com/mozilla/DeepSpeech/releases/latest>`_, but modifying them to use different datasets or language model construction parameters should be simple.
The scripts are geared towards replicating the language model files we release as part of `STT model releases <https://github.com/coqui-ai/STT/releases/latest>`_, but modifying them to use different datasets or language model construction parameters should be simple.
Decoding modes
^^^^^^^^^^^^^^
DeepSpeech currently supports two modes of operation with significant differences at both training and decoding time. Note that Bytes output mode is experimental and has not been tested for languages other than Chinese Mandarin.
🐸STT currently supports two modes of operation with significant differences at both training and decoding time. Note that Bytes output mode is experimental and has not been tested for languages other than Chinese Mandarin.
Default mode (alphabet based)

34
doc/DotNet-API.rst
View File

@ -2,18 +2,18 @@
==============
DeepSpeech Class
STT Class
----------------
.. doxygenclass:: DeepSpeechClient::DeepSpeech
:project: deepspeech-dotnet
.. doxygenclass:: STTClient::STT
:project: stt-dotnet
:members:
DeepSpeechStream Class
Stream Class
----------------------
.. doxygenclass:: DeepSpeechClient::Models::DeepSpeechStream
:project: deepspeech-dotnet
.. doxygenclass:: STTClient::Models::Stream
:project: stt-dotnet
:members:
ErrorCodes
@ -21,33 +21,33 @@ ErrorCodes
See also the main definition including descriptions for each error in :ref:`error-codes`.
.. doxygenenum:: DeepSpeechClient::Enums::ErrorCodes
:project: deepspeech-dotnet
.. doxygenenum:: STTClient::Enums::ErrorCodes
:project: stt-dotnet
Metadata
--------
.. doxygenclass:: DeepSpeechClient::Models::Metadata
:project: deepspeech-dotnet
.. doxygenclass:: STTClient::Models::Metadata
:project: stt-dotnet
:members: Transcripts
CandidateTranscript
-------------------
.. doxygenclass:: DeepSpeechClient::Models::CandidateTranscript
:project: deepspeech-dotnet
.. doxygenclass:: STTClient::Models::CandidateTranscript
:project: stt-dotnet
:members: Tokens, Confidence
TokenMetadata
-------------
.. doxygenclass:: DeepSpeechClient::Models::TokenMetadata
:project: deepspeech-dotnet
.. doxygenclass:: STTClient::Models::TokenMetadata
:project: stt-dotnet
:members: Text, Timestep, StartTime
DeepSpeech Interface
STT Interface
--------------------
.. doxygeninterface:: DeepSpeechClient::Interfaces::IDeepSpeech
:project: deepspeech-dotnet
.. doxygeninterface:: STTClient::Interfaces::ISTT
:project: stt-dotnet
:members:

8
doc/DotNet-Examples.rst
View File

@ -1,12 +1,12 @@
.NET API Usage example
======================
Examples are from `native_client/dotnet/DeepSpeechConsole/Program.cs`.
Examples are from `native_client/dotnet/STTConsole/Program.cs`.
Creating a model instance and loading model
-------------------------------------------
.. literalinclude:: ../native_client/dotnet/DeepSpeechConsole/Program.cs
.. literalinclude:: ../native_client/dotnet/STTConsole/Program.cs
:language: csharp
:linenos:
:lineno-match:
@ -16,7 +16,7 @@ Creating a model instance and loading model
Performing inference
--------------------
.. literalinclude:: ../native_client/dotnet/DeepSpeechConsole/Program.cs
.. literalinclude:: ../native_client/dotnet/STTConsole/Program.cs
:language: csharp
:linenos:
:lineno-match:
@ -26,4 +26,4 @@ Performing inference
Full source code
----------------
See :download:`Full source code<../native_client/dotnet/DeepSpeechConsole/Program.cs>`.
See :download:`Full source code<../native_client/dotnet/STTConsole/Program.cs>`.

2
doc/Error-Codes.rst
View File

@ -5,7 +5,7 @@ Error codes
Below is the definition for all error codes used in the API, their numerical values, and a human readable description.
.. literalinclude:: ../native_client/deepspeech.h
.. literalinclude:: ../native_client/coqui-stt.h
:language: c
:start-after: sphinx-doc: error_code_listing_start
:end-before: sphinx-doc: error_code_listing_end

4
doc/Flags.rst
View File

@ -3,12 +3,12 @@
Command-line flags for the training scripts
===========================================
Below you can find the definition of all command-line flags supported by the training scripts. This includes ``DeepSpeech.py``, ``evaluate.py``, ``evaluate_tflite.py``, ``transcribe.py`` and ``lm_optimizer.py``.
Below you can find the definition of all command-line flags supported by the training scripts. This includes ``train.py``, ``evaluate.py``, ``evaluate_tflite.py``, ``transcribe.py`` and ``lm_optimizer.py``.
Flags
-----
.. literalinclude:: ../training/deepspeech_training/util/flags.py
.. literalinclude:: ../training/coqui_stt_training/util/flags.py
:language: python
:linenos:
:lineno-match:

8
doc/HotWordBoosting-Examples.rst
View File

@ -1,7 +1,7 @@
Hot-word boosting API Usage example
===================================
With DeepSpeech 0.9 release a new API feature was introduced that allows boosting probability from the scorer of given words. It is exposed in all bindings (C, Python, JS, Java and .Net).
With the 🐸STT 0.9 release a new API feature was introduced that allows boosting probability from the scorer of given words. It is exposed in all bindings (C, Python, JS, Java and .Net).
Currently, it provides three methods for the Model class:
@ -19,11 +19,11 @@ It is worth noting that boosting non-existent words in scorer (mostly proper nou
Adjusting the boosting value
----------------------------
For hot-word boosting it is hard to determine what the optimal value that one might be searching for is. Additionally, this is dependant on the input audio file. In practice, as it was reported by DeepSpeech users, the value should be not bigger than 20.0 for positive value boosting. Nevertheless, each usecase is different and you might need to adjust values on your own.
For hot-word boosting it is hard to determine what the optimal value that one might be searching for is. Additionally, this is dependant on the input audio file. In practice, as it was reported by 🐸STT users, the value should be not bigger than 20.0 for positive value boosting. Nevertheless, each usecase is different and you might need to adjust values on your own.
There is a user contributed script available on ``DeepSpeech-examples`` repository for adjusting boost values:
There is a user contributed script available on ``STT-examples`` repository for adjusting boost values:
`https://github.com/mozilla/DeepSpeech-examples/tree/master/hotword_adjusting <https://github.com/mozilla/DeepSpeech-examples/tree/master/hotword_adjusting>`_.
`https://github.com/coqui-ai/STT-examples/tree/master/hotword_adjusting <https://github.com/coqui-ai/STT-examples/tree/master/hotword_adjusting>`_.
Positive value boosting

18
doc/Java-API.rst
View File

@ -1,29 +1,29 @@
Java
====
DeepSpeechModel
STTModel
---------------
.. doxygenclass:: org::deepspeech::libdeepspeech::DeepSpeechModel
:project: deepspeech-java
.. doxygenclass:: ai::coqui::libstt::STTModel
:project: stt-java
:members:
Metadata
--------
.. doxygenclass:: org::deepspeech::libdeepspeech::Metadata
:project: deepspeech-java
.. doxygenclass:: ai::coqui::libstt::Metadata
:project: stt-java
:members: getNumTranscripts, getTranscript
CandidateTranscript
-------------------
.. doxygenclass:: org::deepspeech::libdeepspeech::CandidateTranscript
:project: deepspeech-java
.. doxygenclass:: ai::coqui::libstt::CandidateTranscript
:project: stt-java
:members: getNumTokens, getConfidence, getToken
TokenMetadata
-------------
.. doxygenclass:: org::deepspeech::libdeepspeech::TokenMetadata
:project: deepspeech-java
.. doxygenclass:: ai::coqui::libstt::TokenMetadata
:project: stt-java
:members: getText, getTimestep, getStartTime

8
doc/Java-Examples.rst
View File

@ -1,12 +1,12 @@
Java API Usage example
======================
Examples are from `native_client/java/app/src/main/java/org/deepspeech/DeepSpeechActivity.java`.
Examples are from `native_client/java/app/src/main/java/ai/coqui/STTActivity.java`.
Creating a model instance and loading model
-------------------------------------------
.. literalinclude:: ../native_client/java/app/src/main/java/org/deepspeech/DeepSpeechActivity.java
.. literalinclude:: ../native_client/java/app/src/main/java/ai/coqui/STTActivity.java
:language: java
:linenos:
:lineno-match:
@ -16,7 +16,7 @@ Creating a model instance and loading model
Performing inference
--------------------
.. literalinclude:: ../native_client/java/app/src/main/java/org/deepspeech/DeepSpeechActivity.java
.. literalinclude:: ../native_client/java/app/src/main/java/ai/coqui/STTActivity.java
:language: java
:linenos:
:lineno-match:
@ -26,4 +26,4 @@ Performing inference
Full source code
----------------
See :download:`Full source code<../native_client/java/app/src/main/java/org/deepspeech/DeepSpeechActivity.java>`.
See :download:`Full source code<../native_client/java/app/src/main/java/ai/coqui/STTActivity.java>`.

2
doc/Makefile
View File

@ -4,7 +4,7 @@
# You can set these variables from the command line.
SPHINXOPTS =
SPHINXBUILD = sphinx-build
SPHINXPROJ = DeepSpeech
SPHINXPROJ = "Coqui STT"
SOURCEDIR = .
BUILDDIR = .build

2
doc/ParallelOptimization.rst
View File

@ -1,7 +1,7 @@
Parallel Optimization
=====================
This is how we implement optimization of the DeepSpeech model across GPUs on a
This is how we implement optimization of the 🐸STT model across GPUs on a
single host. Parallel optimization can take on various forms. For example
one can use asynchronous updates of the model, synchronous updates of the model,
or some combination of the two.

28
doc/SUPPORTED_PLATFORMS.rst
View File

@ -9,61 +9,61 @@ Linux / AMD64 without GPU
^^^^^^^^^^^^^^^^^^^^^^^^^
* x86-64 CPU with AVX/FMA (one can rebuild without AVX/FMA, but it might slow down inference)
* Ubuntu 14.04+ (glibc >= 2.19, libstdc++6 >= 4.8)
* Full TensorFlow runtime (``deepspeech`` packages)
* TensorFlow Lite runtime (``deepspeech-tflite`` packages)
* Full TensorFlow runtime (``stt`` packages)
* TensorFlow Lite runtime (``stt-tflite`` packages)
Linux / AMD64 with GPU
^^^^^^^^^^^^^^^^^^^^^^
* x86-64 CPU with AVX/FMA (one can rebuild without AVX/FMA, but it might slow down inference)
* Ubuntu 14.04+ (glibc >= 2.19, libstdc++6 >= 4.8)
* CUDA 10.0 (and capable GPU)
* Full TensorFlow runtime (``deepspeech`` packages)
* TensorFlow Lite runtime (``deepspeech-tflite`` packages)
* Full TensorFlow runtime (``stt`` packages)
* TensorFlow Lite runtime (``stt-tflite`` packages)
Linux / ARMv7
^^^^^^^^^^^^^
* Cortex-A53 compatible ARMv7 SoC with Neon support
* Raspbian Buster-compatible distribution
* TensorFlow Lite runtime (``deepspeech-tflite`` packages)
* TensorFlow Lite runtime (``stt-tflite`` packages)
Linux / Aarch64
^^^^^^^^^^^^^^^
* Cortex-A72 compatible Aarch64 SoC
* ARMbian Buster-compatible distribution
* TensorFlow Lite runtime (``deepspeech-tflite`` packages)
* TensorFlow Lite runtime (``stt-tflite`` packages)
Android / ARMv7
^^^^^^^^^^^^^^^
* ARMv7 SoC with Neon support
* Android 7.0-10.0
* NDK API level >= 21
* TensorFlow Lite runtime (``deepspeech-tflite`` packages)
* TensorFlow Lite runtime (``stt-tflite`` packages)
Android / Aarch64
^^^^^^^^^^^^^^^^^
* Aarch64 SoC
* Android 7.0-10.0
* NDK API level >= 21
* TensorFlow Lite runtime (``deepspeech-tflite`` packages)
* TensorFlow Lite runtime (``stt-tflite`` packages)
macOS / AMD64
^^^^^^^^^^^^^
* x86-64 CPU with AVX/FMA (one can rebuild without AVX/FMA, but it might slow down inference)
* macOS >= 10.10
* Full TensorFlow runtime (``deepspeech`` packages)
* TensorFlow Lite runtime (``deepspeech-tflite`` packages)
* Full TensorFlow runtime (``stt`` packages)
* TensorFlow Lite runtime (``stt-tflite`` packages)
Windows / AMD64 without GPU
^^^^^^^^^^^^^^^^^^^^^^^^^^^
* x86-64 CPU with AVX/FMA (one can rebuild without AVX/FMA, but it might slow down inference)
* Windows Server >= 2012 R2 ; Windows >= 8.1
* Full TensorFlow runtime (``deepspeech`` packages)
* TensorFlow Lite runtime (``deepspeech-tflite`` packages)
* Full TensorFlow runtime (``stt`` packages)
* TensorFlow Lite runtime (``stt-tflite`` packages)
Windows / AMD64 with GPU
^^^^^^^^^^^^^^^^^^^^^^^^
* x86-64 CPU with AVX/FMA (one can rebuild without AVX/FMA, but it might slow down inference)
* Windows Server >= 2012 R2 ; Windows >= 8.1
* CUDA 10.0 (and capable GPU)
* Full TensorFlow runtime (``deepspeech`` packages)
* TensorFlow Lite runtime (``deepspeech-tflite`` packages)
* Full TensorFlow runtime (``stt`` packages)
* TensorFlow Lite runtime (``stt-tflite`` packages)

10
doc/Scorer.rst
View File

@ -3,11 +3,11 @@
External scorer scripts
=======================
DeepSpeech pre-trained models include an external scorer. This document explains how to reproduce our external scorer, as well as adapt the scripts to create your own.
🐸STT pre-trained models include an external scorer. This document explains how to reproduce our external scorer, as well as adapt the scripts to create your own.
The scorer is composed of two sub-components, a KenLM language model and a trie data structure containing all words in the vocabulary. In order to create the scorer package, first we must create a KenLM language model (using ``data/lm/generate_lm.py``, and then use ``generate_scorer_package`` to create the final package file including the trie data structure.
The ``generate_scorer_package`` binary is part of the native client package that is included with official releases. You can find the appropriate archive for your platform in the `GitHub release downloads <https://github.com/mozilla/DeepSpeech/releases/latest>`_. The native client package is named ``native_client.{arch}.{config}.{plat}.tar.xz``, where ``{arch}`` is the architecture the binary was built for, for example ``amd64`` or ``arm64``, ``config`` is the build configuration, which for building decoder packages does not matter, and ``{plat}`` is the platform the binary was built-for, for example ``linux`` or ``osx``. If you wanted to run the ``generate_scorer_package`` binary on a Linux desktop, you would download ``native_client.amd64.cpu.linux.tar.xz``.
The ``generate_scorer_package`` binary is part of the native client package that is included with official releases. You can find the appropriate archive for your platform in the `GitHub release downloads <https://github.com/coqui-ai/STT/releases/latest>`_. The native client package is named ``native_client.{arch}.{config}.{plat}.tar.xz``, where ``{arch}`` is the architecture the binary was built for, for example ``amd64`` or ``arm64``, ``config`` is the build configuration, which for building decoder packages does not matter, and ``{plat}`` is the platform the binary was built-for, for example ``linux`` or ``osx``. If you wanted to run the ``generate_scorer_package`` binary on a Linux desktop, you would download ``native_client.amd64.cpu.linux.tar.xz``.
Reproducing our external scorer
-------------------------------
@ -26,7 +26,7 @@ Then use the ``generate_lm.py`` script to generate ``lm.binary`` and ``vocab-500
As input you can use a plain text (e.g. ``file.txt``) or gzipped (e.g. ``file.txt.gz``) text file with one sentence in each line.
If you are using a container created from ``Dockerfile.build``, you can use ``--kenlm_bins /DeepSpeech/native_client/kenlm/build/bin/``.
If you are using a container created from ``Dockerfile.build``, you can use ``--kenlm_bins /STT/native_client/kenlm/build/bin/``.
Else you have to build `KenLM <https://github.com/kpu/kenlm>`_ first and then pass the build directory to the script.
.. code-block:: bash
@ -44,7 +44,7 @@ Afterwards you can use ``generate_scorer_package`` to generate the scorer packag
cd data/lm
# Download and extract appropriate native_client package:
curl -LO http://github.com/mozilla/DeepSpeech/releases/...
curl -LO http://github.com/coqui-ai/STT/releases/...
tar xvf native_client.*.tar.xz
./generate_scorer_package --alphabet ../alphabet.txt --lm lm.binary --vocab vocab-500000.txt \
--package kenlm.scorer --default_alpha 0.931289039105002 --default_beta 1.1834137581510284
@ -59,6 +59,6 @@ Building your own scorer can be useful if you're using models in a narrow usage
The LibriSpeech LM training text used by our scorer is around 4GB uncompressed, which should give an idea of the size of a corpus needed for a reasonable language model for general speech recognition. For more constrained use cases with smaller vocabularies, you don't need as much data, but you should still try to gather as much as you can.
With a text corpus in hand, you can then re-use ``generate_lm.py`` and ``generate_scorer_package`` to create your own scorer that is compatible with DeepSpeech clients and language bindings. Before building the language model, you must first familiarize yourself with the `KenLM toolkit <https://kheafield.com/code/kenlm/>`_. Most of the options exposed by the ``generate_lm.py`` script are simply forwarded to KenLM options of the same name, so you must read the KenLM documentation in order to fully understand their behavior.
With a text corpus in hand, you can then re-use ``generate_lm.py`` and ``generate_scorer_package`` to create your own scorer that is compatible with 🐸STT clients and language bindings. Before building the language model, you must first familiarize yourself with the `KenLM toolkit <https://kheafield.com/code/kenlm/>`_. Most of the options exposed by the ``generate_lm.py`` script are simply forwarded to KenLM options of the same name, so you must read the KenLM documentation in order to fully understand their behavior.
After using ``generate_lm.py`` to create a KenLM language model binary file, you can use ``generate_scorer_package`` to create a scorer package as described in the previous section. Note that we have a :github:`lm_optimizer.py script <lm_optimizer.py>` which can be used to find good default values for alpha and beta. To use it, you must first generate a package with any value set for default alpha and beta flags. For this step, it doesn't matter what values you use, as they'll be overridden by ``lm_optimizer.py`` later. Then, use ``lm_optimizer.py`` with this scorer file to find good alpha and beta values. Finally, use ``generate_scorer_package`` again, this time with the new values.

6
doc/Structs.rst
View File

@ -5,19 +5,19 @@ Metadata
--------
.. doxygenstruct:: Metadata
:project: deepspeech-c
:project: stt-c
:members:
CandidateTranscript
-------------------
.. doxygenstruct:: CandidateTranscript
:project: deepspeech-c
:project: stt-c
:members:
TokenMetadata
-------------
.. doxygenstruct:: TokenMetadata
:project: deepspeech-c
:project: stt-c
:members:

72
doc/TRAINING.rst
View File

@ -15,11 +15,11 @@ Prerequisites for training a model
Getting the training code
^^^^^^^^^^^^^^^^^^^^^^^^^
Clone the latest released stable branch from Github (e.g. 0.9.3, check `here <https://github.com/mozilla/DeepSpeech/releases>`_):
Clone the latest released stable branch from Github (e.g. 0.9.3, check `here <https://github.com/coqui-ai/STT/releases>`_):
.. code-block:: bash
git clone --branch v0.9.3 https://github.com/mozilla/DeepSpeech
git clone --branch v0.9.3 https://github.com/coqui-ai/STT
If you plan on committing code or you want to report bugs, please use the master branch.
@ -28,31 +28,31 @@ Creating a virtual environment
Throughout the documentation we assume you are using **virtualenv** to manage your Python environments. This setup is the one used and recommended by the project authors and is the easiest way to make sure you won't run into environment issues. If you're using **Anaconda, Miniconda or Mamba**, first read the instructions at :ref:`training-with-conda` and then continue from the installation step below.
In creating a virtual environment you will create a directory containing a ``python3`` binary and everything needed to run deepspeech. You can use whatever directory you want. For the purpose of the documentation, we will rely on ``$HOME/tmp/deepspeech-train-venv``. You can create it using this command:
In creating a virtual environment you will create a directory containing a ``python3`` binary and everything needed to run 🐸STT. You can use whatever directory you want. For the purpose of the documentation, we will rely on ``$HOME/tmp/coqui-stt-train-venv``. You can create it using this command:
.. code-block::
$ python3 -m venv $HOME/tmp/deepspeech-train-venv/
$ python3 -m venv $HOME/tmp/coqui-stt-train-venv/
Once this command completes successfully, the environment will be ready to be activated.
Activating the environment
^^^^^^^^^^^^^^^^^^^^^^^^^^
Each time you need to work with DeepSpeech, you have to *activate* this virtual environment. This is done with this simple command:
Each time you need to work with 🐸STT, you have to *activate* this virtual environment. This is done with this simple command:
.. code-block::
$ source $HOME/tmp/deepspeech-train-venv/bin/activate
$ source $HOME/tmp/coqui-stt-train-venv/bin/activate
Installing DeepSpeech Training Code and its dependencies
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Installing Coqui STT Training Code and its dependencies
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Install the required dependencies using ``pip3``\ :
.. code-block:: bash
cd DeepSpeech
cd STT
pip3 install --upgrade pip==20.2.2 wheel==0.34.2 setuptools==49.6.0
pip3 install --upgrade -e .
@ -95,11 +95,11 @@ This should ensure that you'll re-use the upstream Python 3 TensorFlow GPU-enabl
make Dockerfile.train
If you want to specify a different DeepSpeech repository / branch, you can pass ``DEEPSPEECH_REPO`` or ``DEEPSPEECH_SHA`` parameters:
If you want to specify a different 🐸STT repository / branch, you can pass ``STT_REPO`` or ``STT_SHA`` parameters:
.. code-block:: bash
make Dockerfile.train DEEPSPEECH_REPO=git://your/fork DEEPSPEECH_SHA=origin/your-branch
make Dockerfile.train STT_REPO=git://your/fork STT_SHA=origin/your-branch
Common Voice training data
^^^^^^^^^^^^^^^^^^^^^^^^^^
@ -112,7 +112,7 @@ After extraction of such a data set, you'll find the following contents:
* the ``*.tsv`` files output by CorporaCreator for the downloaded language
* the mp3 audio files they reference in a ``clips`` sub-directory.
For bringing this data into a form that DeepSpeech understands, you have to run the CommonVoice v2.0 importer (\ ``bin/import_cv2.py``\ ):
For bringing this data into a form that 🐸STT understands, you have to run the CommonVoice v2.0 importer (\ ``bin/import_cv2.py``\ ):
.. code-block:: bash
@ -134,22 +134,22 @@ The CSV files comprise of the following fields:
* ``wav_filesize`` - samples size given in bytes, used for sorting the data before training. Expects integer.
* ``transcript`` - transcription target for the sample.
To use Common Voice data during training, validation and testing, you pass (comma separated combinations of) their filenames into ``--train_files``\ , ``--dev_files``\ , ``--test_files`` parameters of ``DeepSpeech.py``.
To use Common Voice data during training, validation and testing, you pass (comma separated combinations of) their filenames into ``--train_files``\ , ``--dev_files``\ , ``--test_files`` parameters of ``train.py``.
If, for example, Common Voice language ``en`` was extracted to ``../data/CV/en/``\ , ``DeepSpeech.py`` could be called like this:
If, for example, Common Voice language ``en`` was extracted to ``../data/CV/en/``\ , ``train.py`` could be called like this:
.. code-block:: bash
python3 DeepSpeech.py --train_files ../data/CV/en/clips/train.csv --dev_files ../data/CV/en/clips/dev.csv --test_files ../data/CV/en/clips/test.csv
python3 train.py --train_files ../data/CV/en/clips/train.csv --dev_files ../data/CV/en/clips/dev.csv --test_files ../data/CV/en/clips/test.csv
Training a model
^^^^^^^^^^^^^^^^
The central (Python) script is ``DeepSpeech.py`` in the project's root directory. For its list of command line options, you can call:
The central (Python) script is ``train.py`` in the project's root directory. For its list of command line options, you can call:
.. code-block:: bash
python3 DeepSpeech.py --helpfull
python3 train.py --helpfull
To get the output of this in a slightly better-formatted way, you can also look at the flag definitions in :ref:`training-flags`.
@ -157,7 +157,7 @@ For executing pre-configured training scenarios, there is a collection of conven
**If you experience GPU OOM errors while training, try reducing the batch size with the ``--train_batch_size``\ , ``--dev_batch_size`` and ``--test_batch_size`` parameters.**
As a simple first example you can open a terminal, change to the directory of the DeepSpeech checkout, activate the virtualenv created above, and run:
As a simple first example you can open a terminal, change to the directory of the 🐸STT checkout, activate the virtualenv created above, and run:
.. code-block:: bash
@ -165,9 +165,9 @@ As a simple first example you can open a terminal, change to the directory of th
This script will train on a small sample dataset composed of just a single audio file, the sample file for the `TIMIT Acoustic-Phonetic Continuous Speech Corpus <https://catalog.ldc.upenn.edu/LDC93S1>`_, which can be overfitted on a GPU in a few minutes for demonstration purposes. From here, you can alter any variables with regards to what dataset is used, how many training iterations are run and the default values of the network parameters.
Feel also free to pass additional (or overriding) ``DeepSpeech.py`` parameters to these scripts. Then, just run the script to train the modified network.
Feel also free to pass additional (or overriding) ``train.py`` parameters to these scripts. Then, just run the script to train the modified network.
Each dataset has a corresponding importer script in ``bin/`` that can be used to download (if it's freely available) and preprocess the dataset. See ``bin/import_librivox.py`` for an example of how to import and preprocess a large dataset for training with DeepSpeech.
Each dataset has a corresponding importer script in ``bin/`` that can be used to download (if it's freely available) and preprocess the dataset. See ``bin/import_librivox.py`` for an example of how to import and preprocess a large dataset for training with 🐸STT.
Some importers might require additional code to properly handled your locale-specific requirements. Such handling is dealt with ``--validate_label_locale`` flag that allows you to source out-of-tree Python script that defines a ``validate_label`` function. Please refer to ``util/importers.py`` for implementation example of that function.
If you don't provide this argument, the default ``validate_label`` function will be used. This one is only intended for English language, so you might have consistency issues in your data for other languages.
@ -191,10 +191,10 @@ Automatic Mixed Precision (AMP) training on GPU for TensorFlow has been recently
Mixed precision training makes use of both FP32 and FP16 precisions where appropriate. FP16 operations can leverage the Tensor cores on NVIDIA GPUs (Volta, Turing or newer architectures) for improved throughput. Mixed precision training also often allows larger batch sizes. Automatic mixed precision training can be enabled by including the flag `--automatic_mixed_precision` at training time:
```
python3 DeepSpeech.py --train_files ./train.csv --dev_files ./dev.csv --test_files ./test.csv --automatic_mixed_precision
python3 train.py --train_files ./train.csv --dev_files ./dev.csv --test_files ./test.csv --automatic_mixed_precision
```
On a Volta generation V100 GPU, automatic mixed precision speeds up DeepSpeech training and evaluation by ~30%-40%.
On a Volta generation V100 GPU, automatic mixed precision speeds up 🐸STT training and evaluation by ~30%-40%.
Checkpointing
^^^^^^^^^^^^^
@ -212,7 +212,7 @@ Refer to the :ref:`usage instructions <usage-docs>` for information on running a
Exporting a model for TFLite
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
If you want to experiment with the TF Lite engine, you need to export a model that is compatible with it, then use the ``--export_tflite`` flags. If you already have a trained model, you can re-export it for TFLite by running ``DeepSpeech.py`` again and specifying the same ``checkpoint_dir`` that you used for training, as well as passing ``--export_tflite --export_dir /model/export/destination``. If you changed the alphabet you also need to add the ``--alphabet_config_path my-new-language-alphabet.txt`` flag.
If you want to experiment with the TF Lite engine, you need to export a model that is compatible with it, then use the ``--export_tflite`` flags. If you already have a trained model, you can re-export it for TFLite by running ``train.py`` again and specifying the same ``checkpoint_dir`` that you used for training, as well as passing ``--export_tflite --export_dir /model/export/destination``. If you changed the alphabet you also need to add the ``--alphabet_config_path my-new-language-alphabet.txt`` flag.
Making a mmap-able model for inference
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@ -236,9 +236,9 @@ Upon sucessfull run, it should report about conversion of a non-zero number of n
Continuing training from a release model
----------------------------------------
There are currently two supported approaches to make use of a pre-trained DeepSpeech model: fine-tuning or transfer-learning. Choosing which one to use is a simple decision, and it depends on your target dataset. Does your data use the same alphabet as the release model? If "Yes": fine-tune. If "No" use transfer-learning.
There are currently two supported approaches to make use of a pre-trained 🐸STT model: fine-tuning or transfer-learning. Choosing which one to use is a simple decision, and it depends on your target dataset. Does your data use the same alphabet as the release model? If "Yes": fine-tune. If "No" use transfer-learning.
If your own data uses the *extact* same alphabet as the English release model (i.e. `a-z` plus `'`) then the release model's output layer will match your data, and you can just fine-tune the existing parameters. However, if you want to use a new alphabet (e.g. Cyrillic `а`, `б`, `д`), the output layer of a release DeepSpeech model will *not* match your data. In this case, you should use transfer-learning (i.e. remove the trained model's output layer, and reinitialize a new output layer that matches your target character set.
If your own data uses the *extact* same alphabet as the English release model (i.e. `a-z` plus `'`) then the release model's output layer will match your data, and you can just fine-tune the existing parameters. However, if you want to use a new alphabet (e.g. Cyrillic `а`, `б`, `д`), the output layer of a release 🐸STT model will *not* match your data. In this case, you should use transfer-learning (i.e. remove the trained model's output layer, and reinitialize a new output layer that matches your target character set.
N.B. - If you have access to a pre-trained model which uses UTF-8 bytes at the output layer you can always fine-tune, because any alphabet should be encodable as UTF-8.
@ -247,14 +247,14 @@ N.B. - If you have access to a pre-trained model which uses UTF-8 bytes at the o
Fine-Tuning (same alphabet)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
If you'd like to use one of the pre-trained models to bootstrap your training process (fine tuning), you can do so by using the ``--checkpoint_dir`` flag in ``DeepSpeech.py``. Specify the path where you downloaded the checkpoint from the release, and training will resume from the pre-trained model.
If you'd like to use one of the pre-trained models to bootstrap your training process (fine tuning), you can do so by using the ``--checkpoint_dir`` flag in ``train.py``. Specify the path where you downloaded the checkpoint from the release, and training will resume from the pre-trained model.
For example, if you want to fine tune the entire graph using your own data in ``my-train.csv``\ , ``my-dev.csv`` and ``my-test.csv``\ , for three epochs, you can something like the following, tuning the hyperparameters as needed:
.. code-block:: bash
mkdir fine_tuning_checkpoints
python3 DeepSpeech.py --n_hidden 2048 --checkpoint_dir path/to/checkpoint/folder --epochs 3 --train_files my-train.csv --dev_files my-dev.csv --test_files my_dev.csv --learning_rate 0.0001
python3 train.py --n_hidden 2048 --checkpoint_dir path/to/checkpoint/folder --epochs 3 --train_files my-train.csv --dev_files my-dev.csv --test_files my_dev.csv --learning_rate 0.0001
Notes about the release checkpoints: the released models were trained with ``--n_hidden 2048``\ , so you need to use that same value when initializing from the release models. Since v0.6.0, the release models are also trained with ``--train_cudnn``\ , so you'll need to specify that as well. If you don't have a CUDA compatible GPU, then you can workaround it by using the ``--load_cudnn`` flag. Use ``--helpfull`` to get more information on how the flags work.
@ -270,17 +270,17 @@ If you try to load a release model without following these steps, you'll get an
Transfer-Learning (new alphabet)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
If you want to continue training an alphabet-based DeepSpeech model (i.e. not a UTF-8 model) on a new language, or if you just want to add new characters to your custom alphabet, you will probably want to use transfer-learning instead of fine-tuning. If you're starting with a pre-trained UTF-8 model -- even if your data comes from a different language or uses a different alphabet -- the model will be able to predict your new transcripts, and you should use fine-tuning instead.
If you want to continue training an alphabet-based 🐸STT model (i.e. not a UTF-8 model) on a new language, or if you just want to add new characters to your custom alphabet, you will probably want to use transfer-learning instead of fine-tuning. If you're starting with a pre-trained UTF-8 model -- even if your data comes from a different language or uses a different alphabet -- the model will be able to predict your new transcripts, and you should use fine-tuning instead.
In a nutshell, DeepSpeech's transfer-learning allows you to remove certain layers from a pre-trained model, initialize new layers for your target data, stitch together the old and new layers, and update all layers via gradient descent. You will remove the pre-trained output layer (and optionally more layers) and reinitialize parameters to fit your target alphabet. The simplest case of transfer-learning is when you remove just the output layer.
In a nutshell, 🐸STT's transfer-learning allows you to remove certain layers from a pre-trained model, initialize new layers for your target data, stitch together the old and new layers, and update all layers via gradient descent. You will remove the pre-trained output layer (and optionally more layers) and reinitialize parameters to fit your target alphabet. The simplest case of transfer-learning is when you remove just the output layer.
In DeepSpeech's implementation of transfer-learning, all removed layers will be contiguous, starting from the output layer. The key flag you will want to experiment with is ``--drop_source_layers``. This flag accepts an integer from ``1`` to ``5`` and allows you to specify how many layers you want to remove from the pre-trained model. For example, if you supplied ``--drop_source_layers 3``, you will drop the last three layers of the pre-trained model: the output layer, penultimate layer, and LSTM layer. All dropped layers will be reinintialized, and (crucially) the output layer will be defined to match your supplied target alphabet.
In 🐸STT's implementation of transfer-learning, all removed layers will be contiguous, starting from the output layer. The key flag you will want to experiment with is ``--drop_source_layers``. This flag accepts an integer from ``1`` to ``5`` and allows you to specify how many layers you want to remove from the pre-trained model. For example, if you supplied ``--drop_source_layers 3``, you will drop the last three layers of the pre-trained model: the output layer, penultimate layer, and LSTM layer. All dropped layers will be reinintialized, and (crucially) the output layer will be defined to match your supplied target alphabet.
You need to specify the location of the pre-trained model with ``--load_checkpoint_dir`` and define where your new model checkpoints will be saved with ``--save_checkpoint_dir``. You need to specify how many layers to remove (aka "drop") from the pre-trained model: ``--drop_source_layers``. You also need to supply your new alphabet file using the standard ``--alphabet_config_path`` (remember, using a new alphabet is the whole reason you want to use transfer-learning).
.. code-block:: bash
python3 DeepSpeech.py \
python3 train.py \
--drop_source_layers 1 \
--alphabet_config_path my-new-language-alphabet.txt \
--save_checkpoint_dir path/to/output-checkpoint/folder \
@ -292,7 +292,7 @@ You need to specify the location of the pre-trained model with ``--load_checkpoi
UTF-8 mode
^^^^^^^^^^
DeepSpeech includes a UTF-8 operating mode which can be useful to model languages with very large alphabets, such as Chinese Mandarin. For details on how it works and how to use it, see :ref:`decoder-docs`.
🐸STT includes a UTF-8 operating mode which can be useful to model languages with very large alphabets, such as Chinese Mandarin. For details on how it works and how to use it, see :ref:`decoder-docs`.
.. _training-data-augmentation:
@ -314,7 +314,7 @@ For example, for the ``overlay`` augmentation:
.. code-block::
python3 DeepSpeech.py --augment overlay[p=0.1,source=/path/to/audio.sdb,snr=20.0] ...
python3 train.py --augment overlay[p=0.1,source=/path/to/audio.sdb,snr=20.0] ...
In the documentation below, whenever a value is specified as ``<float-range>`` or ``<int-range>``, it supports one of the follow formats:
@ -485,7 +485,7 @@ Example training with all augmentations:
.. code-block:: bash
python -u DeepSpeech.py \
python -u train.py \
--train_files "train.sdb" \
--feature_cache ./feature.cache \
--cache_for_epochs 10 \
@ -541,5 +541,5 @@ To prevent common problems, make sure you **always use a separate environment wh
.. code-block:: bash
(base) $ conda create -n deepspeech python=3.7
(base) $ conda activate deepspeech
(base) $ conda create -n coqui-stt python=3.7
(base) $ conda activate coqui-stt

105
doc/USING.rst
View File

@ -3,7 +3,7 @@
Using a Pre-trained Model
=========================
Inference using a DeepSpeech pre-trained model can be done with a client/language binding package. We have four clients/language bindings in this repository, listed below, and also a few community-maintained clients/language bindings in other repositories, listed `further down in this README <#third-party-bindings>`_.
Inference using a 🐸STT pre-trained model can be done with a client/language binding package. We have four clients/language bindings in this repository, listed below, and also a few community-maintained clients/language bindings in other repositories, listed `further down in this README <#third-party-bindings>`_.
* :ref:`The C API <c-usage>`.
* :ref:`The Python package/language binding <py-usage>`
@ -13,7 +13,7 @@ Inference using a DeepSpeech pre-trained model can be done with a client/languag
.. _runtime-deps:
Running ``deepspeech`` might, see below, require some runtime dependencies to be already installed on your system:
Running ``stt`` might, see below, require some runtime dependencies to be already installed on your system:
* ``sox`` - The Python and Node.JS clients use SoX to resample files to 16kHz.
* ``libgomp1`` - libsox (statically linked into the clients) depends on OpenMP. Some people have had to install this manually.
@ -33,23 +33,23 @@ The GPU capable builds (Python, NodeJS, C++, etc) depend on CUDA 10.1 and CuDNN
Getting the pre-trained model
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
If you want to use the pre-trained English model for performing speech-to-text, you can download it (along with other important inference material) from the DeepSpeech `releases page <https://github.com/mozilla/DeepSpeech/releases>`_. Alternatively, you can run the following command to download the model files in your current directory:
If you want to use the pre-trained English model for performing speech-to-text, you can download it (along with other important inference material) from the 🐸STT `releases page <https://github.com/coqui-ai/STT/releases>`_. Alternatively, you can run the following command to download the model files in your current directory:
.. code-block:: bash
wget https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/deepspeech-0.9.3-models.pbmm
wget https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/deepspeech-0.9.3-models.scorer
wget https://github.com/coqui-ai/STT/releases/download/v0.9.3/coqui-stt-0.9.3-models.pbmm
wget https://github.com/coqui-ai/STT/releases/download/v0.9.3/coqui-stt-0.9.3-models.scorer
There are several pre-trained model files available in official releases. Files ending in ``.pbmm`` are compatible with clients and language bindings built against the standard TensorFlow runtime. Usually these packages are simply called ``deepspeech``. These files are also compatible with CUDA enabled clients and language bindings. These packages are usually called ``deepspeech-gpu``. Files ending in ``.tflite`` are compatible with clients and language bindings built against the `TensorFlow Lite runtime <https://www.tensorflow.org/lite/>`_. These models are optimized for size and performance in low power devices. On desktop platforms, the compatible packages are called ``deepspeech-tflite``. On Android and Raspberry Pi, we only publish TensorFlow Lite enabled packages, and they are simply called ``deepspeech``. You can see a full list of supported platforms and which TensorFlow runtime is supported at :ref:`supported-platforms-inference`.
There are several pre-trained model files available in official releases. Files ending in ``.pbmm`` are compatible with clients and language bindings built against the standard TensorFlow runtime. Usually these packages are simply called ``stt``. These files are also compatible with CUDA enabled clients and language bindings. These packages are usually called ``stt-gpu``. Files ending in ``.tflite`` are compatible with clients and language bindings built against the `TensorFlow Lite runtime <https://www.tensorflow.org/lite/>`_. These models are optimized for size and performance in low power devices. On desktop platforms, the compatible packages are called ``stt-tflite``. On Android and Raspberry Pi, we only publish TensorFlow Lite enabled packages, and they are simply called ``stt``. You can see a full list of supported platforms and which TensorFlow runtime is supported at :ref:`supported-platforms-inference`.
+--------------------+---------------------+---------------------+
| Package/Model type | .pbmm | .tflite |
+====================+=====================+=====================+
| deepspeech | Depends on platform | Depends on platform |
| stt | Depends on platform | Depends on platform |
+--------------------+---------------------+---------------------+
| deepspeech-gpu | ✅ | ❌ |
| stt-gpu | ✅ | ❌ |
+--------------------+---------------------+---------------------+
| deepspeech-tflite | ❌ | ✅ |
| stt-tflite | ❌ | ✅ |
+--------------------+---------------------+---------------------+
Finally, the pre-trained model files also include files ending in ``.scorer``. These are external scorers (language models) that are used at inference time in conjunction with an acoustic model (``.pbmm`` or ``.tflite`` file) to produce transcriptions. We also provide further documentation on :ref:`the decoding process <decoder-docs>` and :ref:`how scorers are generated <scorer-scripts>`.
@ -61,82 +61,82 @@ The release notes include detailed information on how the released models were t
The process for training an acoustic model is described in :ref:`training-docs`. In particular, fine tuning a release model using your own data can be a good way to leverage relatively smaller amounts of data that would not be sufficient for training a new model from scratch. See the :ref:`fine tuning and transfer learning sections <training-fine-tuning>` for more information. :ref:`Data augmentation <training-data-augmentation>` can also be a good way to increase the value of smaller training sets.
Creating your own external scorer from text data is another way that you can adapt the model to your specific needs. The process and tools used to generate an external scorer package are described in :ref:`scorer-scripts` and an overview of how the external scorer is used by DeepSpeech to perform inference is available in :ref:`decoder-docs`. Generating a smaller scorer from a single purpose text dataset is a quick process and can bring significant accuracy improvements, specially for more constrained, limited vocabulary applications.
Creating your own external scorer from text data is another way that you can adapt the model to your specific needs. The process and tools used to generate an external scorer package are described in :ref:`scorer-scripts` and an overview of how the external scorer is used by 🐸STT to perform inference is available in :ref:`decoder-docs`. Generating a smaller scorer from a single purpose text dataset is a quick process and can bring significant accuracy improvements, specially for more constrained, limited vocabulary applications.
Model compatibility
^^^^^^^^^^^^^^^^^^^
DeepSpeech models are versioned to keep you from trying to use an incompatible graph with a newer client after a breaking change was made to the code. If you get an error saying your model file version is too old for the client, you should either upgrade to a newer model release, re-export your model from the checkpoint using a newer version of the code, or downgrade your client if you need to use the old model and can't re-export it.
🐸STT models are versioned to keep you from trying to use an incompatible graph with a newer client after a breaking change was made to the code. If you get an error saying your model file version is too old for the client, you should either upgrade to a newer model release, re-export your model from the checkpoint using a newer version of the code, or downgrade your client if you need to use the old model and can't re-export it.
.. _py-usage:
Using the Python package
^^^^^^^^^^^^^^^^^^^^^^^^
Pre-built binaries which can be used for performing inference with a trained model can be installed with ``pip3``. You can then use the ``deepspeech`` binary to do speech-to-text on an audio file:
Pre-built binaries which can be used for performing inference with a trained model can be installed with ``pip3``. You can then use the ``stt`` binary to do speech-to-text on an audio file:
For the Python bindings, it is highly recommended that you perform the installation within a Python 3.5 or later virtual environment. You can find more information about those in `this documentation <http://docs.python-guide.org/en/latest/dev/virtualenvs/>`_.
We will continue under the assumption that you already have your system properly setup to create new virtual environments.
Create a DeepSpeech virtual environment
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Create a Coqui STT virtual environment
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In creating a virtual environment you will create a directory containing a ``python3`` binary and everything needed to run deepspeech. You can use whatever directory you want. For the purpose of the documentation, we will rely on ``$HOME/tmp/deepspeech-venv``. You can create it using this command:
In creating a virtual environment you will create a directory containing a ``python3`` binary and everything needed to run 🐸STT. You can use whatever directory you want. For the purpose of the documentation, we will rely on ``$HOME/tmp/coqui-stt-venv``. You can create it using this command:
.. code-block::
$ virtualenv -p python3 $HOME/tmp/deepspeech-venv/
$ virtualenv -p python3 $HOME/tmp/coqui-stt-venv/
Once this command completes successfully, the environment will be ready to be activated.
Activating the environment
~~~~~~~~~~~~~~~~~~~~~~~~~~
Each time you need to work with DeepSpeech, you have to *activate* this virtual environment. This is done with this simple command:
Each time you need to work with 🐸STT, you have to *activate* this virtual environment. This is done with this simple command:
.. code-block::
$ source $HOME/tmp/deepspeech-venv/bin/activate
$ source $HOME/tmp/coqui-stt-venv/bin/activate
Installing DeepSpeech Python bindings
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Installing Coqui STT Python bindings
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Once your environment has been set-up and loaded, you can use ``pip3`` to manage packages locally. On a fresh setup of the ``virtualenv``\ , you will have to install the DeepSpeech wheel. You can check if ``deepspeech`` is already installed with ``pip3 list``.
Once your environment has been set-up and loaded, you can use ``pip3`` to manage packages locally. On a fresh setup of the ``virtualenv``\ , you will have to install the 🐸STT wheel. You can check if ``stt`` is already installed with ``pip3 list``.
To perform the installation, just use ``pip3`` as such:
.. code-block::
$ pip3 install deepspeech
$ pip3 install stt
If ``deepspeech`` is already installed, you can update it as such:
If ``stt`` is already installed, you can update it as such:
.. code-block::
$ pip3 install --upgrade deepspeech
$ pip3 install --upgrade stt
Alternatively, if you have a supported NVIDIA GPU on Linux, you can install the GPU specific package as follows:
.. code-block::
$ pip3 install deepspeech-gpu
$ pip3 install stt-gpu
See the `release notes <https://github.com/mozilla/DeepSpeech/releases>`_ to find which GPUs are supported. Please ensure you have the required `CUDA dependency <#cuda-dependency>`_.
See the `release notes <https://github.com/coqui-ai/STT/releases>`_ to find which GPUs are supported. Please ensure you have the required `CUDA dependency <#cuda-dependency>`_.
You can update ``deepspeech-gpu`` as follows:
You can update ``stt-gpu`` as follows:
.. code-block::
$ pip3 install --upgrade deepspeech-gpu
$ pip3 install --upgrade stt-gpu
In both cases, ``pip3`` should take care of installing all the required dependencies. After installation has finished, you should be able to call ``deepspeech`` from the command-line.
In both cases, ``pip3`` should take care of installing all the required dependencies. After installation has finished, you should be able to call ``stt`` from the command-line.
Note: the following command assumes you `downloaded the pre-trained model <#getting-the-pre-trained-model>`_.
.. code-block:: bash
deepspeech --model deepspeech-0.9.3-models.pbmm --scorer deepspeech-0.9.3-models.scorer --audio my_audio_file.wav
stt --model stt-0.9.3-models.pbmm --scorer stt-0.9.3-models.scorer --audio my_audio_file.wav
The ``--scorer`` argument is optional, and represents an external language model to be used when transcribing the audio.
@ -151,7 +151,9 @@ You can download the JS bindings using ``npm``\ :
.. code-block:: bash
npm install deepspeech
npm install stt
Special thanks to `Huan - Google Developers Experts in Machine Learning (ML GDE) <https://github.com/huan>`_ for providing the STT project name on npmjs.org
Please note that as of now, we support:
- Node.JS versions 4 to 13.
@ -163,9 +165,9 @@ Alternatively, if you're using Linux and have a supported NVIDIA GPU, you can in
.. code-block:: bash
npm install deepspeech-gpu
npm install stt-gpu
See the `release notes <https://github.com/mozilla/DeepSpeech/releases>`_ to find which GPUs are supported. Please ensure you have the required `CUDA dependency <#cuda-dependency>`_.
See the `release notes <https://github.com/coqui-ai/STT/releases>`_ to find which GPUs are supported. Please ensure you have the required `CUDA dependency <#cuda-dependency>`_.
See the :ref:`TypeScript client <js-api-example>` for an example of how to use the bindings programatically.
@ -174,7 +176,7 @@ See the :ref:`TypeScript client <js-api-example>` for an example of how to use t
Using the command-line client
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
To download the pre-built binaries for the ``deepspeech`` command-line (compiled C++) client, use ``util/taskcluster.py``\ :
To download the pre-built binaries for the ``stt`` command-line (compiled C++) client, use ``util/taskcluster.py``\ :
.. code-block:: bash
@ -186,23 +188,23 @@ or if you're on macOS:
python3 util/taskcluster.py --arch osx --target .
also, if you need some binaries different than current master, like ``v0.2.0-alpha.6``\ , you can use ``--branch``\ :
also, if you need some binaries different than current main branch, like ``v0.2.0-alpha.6``\ , you can use ``--branch``\ :
.. code-block:: bash
python3 util/taskcluster.py --branch "v0.2.0-alpha.6" --target "."
The script ``taskcluster.py`` will download ``native_client.tar.xz`` (which includes the ``deepspeech`` binary and associated libraries) and extract it into the current folder. Also, ``taskcluster.py`` will download binaries for Linux/x86_64 by default, but you can override that behavior with the ``--arch`` parameter. See the help info with ``python util/taskcluster.py -h`` for more details. Specific branches of DeepSpeech or TensorFlow can be specified as well.
The script ``taskcluster.py`` will download ``native_client.tar.xz`` (which includes the ``stt`` binary and associated libraries) and extract it into the current folder. Also, ``taskcluster.py`` will download binaries for Linux/x86_64 by default, but you can override that behavior with the ``--arch`` parameter. See the help info with ``python util/taskcluster.py -h`` for more details. Specific branches of 🐸STT or TensorFlow can be specified as well.
Alternatively you may manually download the ``native_client.tar.xz`` from the [releases](https://github.com/mozilla/DeepSpeech/releases).
Alternatively you may manually download the ``native_client.tar.xz`` from the [releases](https://github.com/coqui-ai/STT/releases).
Note: the following command assumes you `downloaded the pre-trained model <#getting-the-pre-trained-model>`_.
.. code-block:: bash
./deepspeech --model deepspeech-0.9.3-models.pbmm --scorer deepspeech-0.9.3-models.scorer --audio audio_input.wav
./stt --model coqui-stt-0.9.3-models.pbmm --scorer coqui-stt-0.9.3-models.scorer --audio audio_input.wav
See the help output with ``./deepspeech -h`` for more details.
See the help output with ``./stt -h`` for more details.
Installing bindings from source
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@ -212,28 +214,27 @@ If pre-built binaries aren't available for your system, you'll need to install t
Dockerfile for building from source
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
We provide ``Dockerfile.build`` to automatically build ``libdeepspeech.so``, the C++ native client, Python bindings, and KenLM.
We provide ``Dockerfile.build`` to automatically build ``libstt.so``, the C++ native client, Python bindings, and KenLM.
You need to generate the Dockerfile from the template using:
.. code-block:: bash
make Dockerfile.build
If you want to specify a different DeepSpeech repository / branch, you can pass ``DEEPSPEECH_REPO`` or ``DEEPSPEECH_SHA`` parameters:
If you want to specify a different repository / branch, you can pass ``STT_REPO`` or ``STT_SHA`` parameters:
.. code-block:: bash
make Dockerfile.build DEEPSPEECH_REPO=git://your/fork DEEPSPEECH_SHA=origin/your-branch
make Dockerfile.build STT_REPO=git://your/fork STT_SHA=origin/your-branch
Third party bindings
^^^^^^^^^^^^^^^^^^^^
.. Third party bindings
^^^^^^^^^^^^^^^^^^^^
In addition to the bindings above, third party developers have started to provide bindings to other languages:
In addition to the bindings above, third party developers have started to provide bindings to other languages:
* `Asticode <https://github.com/asticode>`_ provides `Golang <https://golang.org>`_ bindings in its `go-astideepspeech <https://github.com/asticode/go-astideepspeech>`_ repo.
* `RustAudio <https://github.com/RustAudio>`_ provide a `Rust <https://www.rust-lang.org>`_ binding, the installation and use of which is described in their `deepspeech-rs <https://github.com/RustAudio/deepspeech-rs>`_ repo.
* `stes <https://github.com/stes>`_ provides preliminary `PKGBUILDs <https://wiki.archlinux.org/index.php/PKGBUILD>`_ to install the client and python bindings on `Arch Linux <https://www.archlinux.org/>`_ in the `arch-deepspeech <https://github.com/stes/arch-deepspeech>`_ repo.
* `gst-deepspeech <https://github.com/Elleo/gst-deepspeech>`_ provides a `GStreamer <https://gstreamer.freedesktop.org/>`_ plugin which can be used from any language with GStreamer bindings.
* `thecodrr <https://github.com/thecodrr>`_ provides `Vlang <https://vlang.io>`_ bindings. The installation and use of which is described in their `vspeech <https://github.com/thecodrr/vspeech>`_ repo.
* `eagledot <https://gitlab.com/eagledot>`_ provides `NIM-lang <https://nim-lang.org/>`_ bindings. The installation and use of which is described in their `nim-deepspeech <https://gitlab.com/eagledot/nim-deepspeech>`_ repo.
* `Asticode <https://github.com/asticode>`_ provides `Golang <https://golang.org>`_ bindings in its `go-astideepspeech <https://github.com/asticode/go-astideepspeech>`_ repo.
* `RustAudio <https://github.com/RustAudio>`_ provide a `Rust <https://www.rust-lang.org>`_ binding, the installation and use of which is described in their `deepspeech-rs <https://github.com/RustAudio/deepspeech-rs>`_ repo.
* `stes <https://github.com/stes>`_ provides preliminary `PKGBUILDs <https://wiki.archlinux.org/index.php/PKGBUILD>`_ to install the client and python bindings on `Arch Linux <https://www.archlinux.org/>`_ in the `arch-deepspeech <https://github.com/stes/arch-deepspeech>`_ repo.
* `gst-deepspeech <https://github.com/Elleo/gst-deepspeech>`_ provides a `GStreamer <https://gstreamer.freedesktop.org/>`_ plugin which can be used from any language with GStreamer bindings.
* `thecodrr <https://github.com/thecodrr>`_ provides `Vlang <https://vlang.io>`_ bindings. The installation and use of which is described in their `vspeech <https://github.com/thecodrr/vspeech>`_ repo.
* `eagledot <https://gitlab.com/eagledot>`_ provides `NIM-lang <https://nim-lang.org/>`_ bindings. The installation and use of which is described in their `nim-deepspeech <https://gitlab.com/eagledot/nim-deepspeech>`_ repo.

32
doc/conf.py
View File

@ -1,6 +1,6 @@
# -*- coding: utf-8 -*-
#
# DeepSpeech documentation build configuration file, created by
# Coqui STT documentation build configuration file, created by
# sphinx-quickstart on Thu Feb 2 21:20:39 2017.
#
# This file is execfile()d with the current directory set to its
@ -24,7 +24,7 @@ import sys
sys.path.insert(0, os.path.abspath('../'))
autodoc_mock_imports = ['deepspeech']
autodoc_mock_imports = ['stt']
# This is in fact only relevant on ReadTheDocs, but we want to run the same way
# on our CI as in RTD to avoid regressions on RTD that we would not catch on
@ -45,9 +45,9 @@ import semver
# -- Project information -----------------------------------------------------
project = u'DeepSpeech'
copyright = '2019-2020 Mozilla Corporation, 2020 DeepSpeech authors'
author = 'DeepSpeech authors'
project = u'Coqui STT'
copyright = '2019-2020 Mozilla Corporation, 2020 DeepSpeech authors, 2021 Coqui GmbH'
author = 'Coqui GmbH'
with open('../VERSION', 'r') as ver:
v = ver.read().strip()
@ -81,9 +81,9 @@ extensions = [
breathe_projects = {
"deepspeech-c": "xml-c/",
"deepspeech-java": "xml-java/",
"deepspeech-dotnet": "xml-dotnet/",
"stt-c": "xml-c/",
"stt-java": "xml-java/",
"stt-dotnet": "xml-dotnet/",
}
js_source_path = "../native_client/javascript/index.ts"
@ -99,7 +99,7 @@ templates_path = ['.templates']
# source_suffix = ['.rst', '.md']
source_suffix = '.rst'
# The master toctree document.
# The main toctree document.
master_doc = 'index'
# The language for content autogenerated by Sphinx. Refer to documentation
@ -147,7 +147,7 @@ html_static_path = ['.static']
# -- Options for HTMLHelp output ------------------------------------------
# Output file base name for HTML help builder.
htmlhelp_basename = 'DeepSpeechdoc'
htmlhelp_basename = 'STTdoc'
# -- Options for LaTeX output ---------------------------------------------
@ -174,8 +174,8 @@ latex_elements = {
# (source start file, target name, title,
# author, documentclass [howto, manual, or own class]).
latex_documents = [
(master_doc, 'DeepSpeech.tex', u'DeepSpeech Documentation',
u'DeepSpeech authors', 'manual'),
(master_doc, 'STT.tex', u'Coqui STT Documentation',
u'Coqui GmbH', 'manual'),
]
@ -184,7 +184,7 @@ latex_documents = [
# One entry per manual page. List of tuples
# (source start file, name, description, authors, manual section).
man_pages = [
(master_doc, 'deepspeech', u'DeepSpeech Documentation',
(master_doc, 'stt', u'Coqui STT Documentation',
[author], 1)
]
@ -195,8 +195,8 @@ man_pages = [
# (source start file, target name, title, author,
# dir menu entry, description, category)
texinfo_documents = [
(master_doc, 'DeepSpeech', u'DeepSpeech Documentation',
author, 'DeepSpeech', 'One line description of project.',
(master_doc, 'STT', u'Coqui STT Documentation',
author, 'STT', 'One line description of project.',
'Miscellaneous'),
]
@ -206,5 +206,5 @@ texinfo_documents = [
# Example configuration for intersphinx: refer to the Python standard library.
intersphinx_mapping = {'https://docs.python.org/': None}
extlinks = {'github': ('https://github.com/mozilla/DeepSpeech/blob/v{}/%s'.format(release),
extlinks = {'github': ('https://github.com/coqui-ai/STT/blob/v{}/%s'.format(release),
'%s')}

2
doc/doxygen-c.conf
View File

@ -790,7 +790,7 @@ WARN_LOGFILE =
# spaces. See also FILE_PATTERNS and EXTENSION_MAPPING
# Note: If this tag is empty the current directory is searched.
INPUT = native_client/deepspeech.h
INPUT = native_client/coqui-stt.h
# This tag can be used to specify the character encoding of the source files
# that doxygen parses. Internally doxygen uses the UTF-8 encoding. Doxygen uses

2
doc/doxygen-dotnet.conf
View File

@ -790,7 +790,7 @@ WARN_LOGFILE =
# spaces. See also FILE_PATTERNS and EXTENSION_MAPPING
# Note: If this tag is empty the current directory is searched.
INPUT = native_client/dotnet/DeepSpeechClient/ native_client/dotnet/DeepSpeechClient/Interfaces/ native_client/dotnet/DeepSpeechClient/Enums/ native_client/dotnet/DeepSpeechClient/Models/
INPUT = native_client/dotnet/STTClient/ native_client/dotnet/STTClient/Interfaces/ native_client/dotnet/STTClient/Enums/ native_client/dotnet/STTClient/Models/
# This tag can be used to specify the character encoding of the source files
# that doxygen parses. Internally doxygen uses the UTF-8 encoding. Doxygen uses

2
doc/doxygen-java.conf
View File

@ -790,7 +790,7 @@ WARN_LOGFILE =
# spaces. See also FILE_PATTERNS and EXTENSION_MAPPING
# Note: If this tag is empty the current directory is searched.
INPUT = native_client/java/libdeepspeech/src/main/java/org/deepspeech/libdeepspeech/ native_client/java/libdeepspeech/src/main/java/org/deepspeech/libdeepspeech_doc/
INPUT = native_client/java/libstt/src/main/java/ai/coqui/libstt/ native_client/java/libstt/src/main/java/ai/coqui/libstt_doc/
# This tag can be used to specify the character encoding of the source files
# that doxygen parses. Internally doxygen uses the UTF-8 encoding. Doxygen uses

44
doc/index.rst
View File

@ -1,54 +1,54 @@
.. DeepSpeech documentation master file, created by
.. Coqui STT documentation main file, created by
sphinx-quickstart on Thu Feb 2 21:20:39 2017.
You can adapt this file completely to your liking, but it should at least
contain the root `toctree` directive.
Welcome to DeepSpeech's documentation!
======================================
Coqui STT
=========
DeepSpeech is an open source Speech-To-Text engine, using a model trained by machine learning techniques based on `Baidu's Deep Speech research paper <https://arxiv.org/abs/1412.5567>`_. Project DeepSpeech uses Google's `TensorFlow <https://www.tensorflow.org/>`_ to make the implementation easier.
Coqui STT (🐸STT) is an open source Speech-To-Text engine, using a model trained by machine learning techniques based on `Baidu's Deep Speech research paper <https://arxiv.org/abs/1412.5567>`_. 🐸STT uses Google's `TensorFlow <https://www.tensorflow.org/>`_ to make the implementation easier.
To install and use DeepSpeech all you have to do is:
To install and use 🐸STT all you have to do is:
.. code-block:: bash
# Create and activate a virtualenv
virtualenv -p python3 $HOME/tmp/deepspeech-venv/
source $HOME/tmp/deepspeech-venv/bin/activate
virtualenv -p python3 $HOME/tmp/stt/
source $HOME/tmp/stt/bin/activate
# Install DeepSpeech
pip3 install deepspeech
# Install 🐸STT
pip3 install stt
# Download pre-trained English model files
curl -LO https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/deepspeech-0.9.3-models.pbmm
curl -LO https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/deepspeech-0.9.3-models.scorer
curl -LO https://github.com/coqui-ai/STT/releases/download/v0.9.3/coqui-stt-0.9.3-models.pbmm
curl -LO https://github.com/coqui-ai/STT/releases/download/v0.9.3/coqui-stt-0.9.3-models.scorer
# Download example audio files
curl -LO https://github.com/mozilla/DeepSpeech/releases/download/v0.9.3/audio-0.9.3.tar.gz
curl -LO https://github.com/coqui-ai/STT/releases/download/v0.9.3/audio-0.9.3.tar.gz
tar xvf audio-0.9.3.tar.gz
# Transcribe an audio file
deepspeech --model deepspeech-0.9.3-models.pbmm --scorer deepspeech-0.9.3-models.scorer --audio audio/2830-3980-0043.wav
stt --model coqui-stt-0.9.3-models.pbmm --scorer coqui-stt-0.9.3-models.scorer --audio audio/2830-3980-0043.wav
A pre-trained English model is available for use and can be downloaded following the instructions in :ref:`the usage docs <usage-docs>`. For the latest release, including pre-trained models and checkpoints, `see the GitHub releases page <https://github.com/mozilla/DeepSpeech/releases/latest>`_.
A pre-trained English model is available for use and can be downloaded following the instructions in :ref:`the usage docs <usage-docs>`. For the latest release, including pre-trained models and checkpoints, `see the GitHub releases page <https://github.com/coqui-ai/STT/releases/latest>`_.
Quicker inference can be performed using a supported NVIDIA GPU on Linux. See the `release notes <https://github.com/mozilla/DeepSpeech/releases/latest>`_ to find which GPUs are supported. To run ``deepspeech`` on a GPU, install the GPU specific package:
Quicker inference can be performed using a supported NVIDIA GPU on Linux. See the `release notes <https://github.com/coqui-ai/STT/releases/latest>`_ to find which GPUs are supported. To run ``stt`` on a GPU, install the GPU specific package:
.. code-block:: bash
# Create and activate a virtualenv
virtualenv -p python3 $HOME/tmp/deepspeech-gpu-venv/
source $HOME/tmp/deepspeech-gpu-venv/bin/activate
virtualenv -p python3 $HOME/tmp/coqui-stt-gpu-venv/
source $HOME/tmp/coqui-stt-gpu-venv/bin/activate
# Install DeepSpeech CUDA enabled package
pip3 install deepspeech-gpu
# Install 🐸STT CUDA enabled package
pip3 install stt-gpu
# Transcribe an audio file.
deepspeech --model deepspeech-0.9.3-models.pbmm --scorer deepspeech-0.9.3-models.scorer --audio audio/2830-3980-0043.wav
stt --model coqui-stt-0.9.3-models.pbmm --scorer coqui-stt-0.9.3-models.scorer --audio audio/2830-3980-0043.wav
Please ensure you have the required :ref:`CUDA dependencies <cuda-inference-deps>`.
See the output of ``deepspeech -h`` for more information on the use of ``deepspeech``. (If you experience problems running ``deepspeech``, please check :ref:`required runtime dependencies <runtime-deps>`).
See the output of ``stt -h`` for more information on the use of ``stt``. (If you experience problems running ``stt``, please check :ref:`required runtime dependencies <runtime-deps>`).
.. toctree::
:maxdepth: 2
@ -78,7 +78,7 @@ See the output of ``deepspeech -h`` for more information on the use of ``deepspe
:maxdepth: 2
:caption: Architecture and training
DeepSpeech
Architecture
Geometry

2
doc/make.bat
View File

@ -9,7 +9,7 @@ if "%SPHINXBUILD%" == "" (
)
set SOURCEDIR=.
set BUILDDIR=.build
set SPHINXPROJ=DeepSpeech
set SPHINXPROJ="Coqui STT"
if "%1" == "" goto help

4
ds_lib.supp
View File

@ -1,10 +1,10 @@
{
deepspeech_tflite_error_reporter
stt_tflite_error_reporter
Memcheck:Leak
match-leak-kinds: reachable
fun:_Znwm
fun:_ZN6tflite20DefaultErrorReporterEv
fun:_ZN16TFLiteModelState4initEPKc
fun:DS_CreateModel
fun:STT_CreateModel
fun:main
}

4
ds_openfst.supp
View File

@ -815,7 +815,7 @@
fun:_ZN6Scorer9load_trieERSt14basic_ifstreamIcSt11char_traitsIcEERKNSt7__cxx1112basic_stringIcS2_SaIcEEE
fun:_ZN6Scorer7load_lmERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE
fun:_ZN6Scorer4initERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEERK8Alphabet
fun:DS_EnableExternalScorer
fun:STT_EnableExternalScorer
fun:main
}
{
@ -831,7 +831,7 @@
fun:_ZN6Scorer9load_trieERSt14basic_ifstreamIcSt11char_traitsIcEERKNSt7__cxx1112basic_stringIcS2_SaIcEEE
fun:_ZN6Scorer7load_lmERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE
fun:_ZN6Scorer4initERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEERK8Alphabet
fun:DS_EnableExternalScorer
fun:STT_EnableExternalScorer
fun:main
}
{

2
evaluate.py
View File

@ -4,7 +4,7 @@ from __future__ import absolute_import, division, print_function
if __name__ == '__main__':
try:
from deepspeech_training import evaluate as ds_evaluate
from coqui_stt_training import evaluate as ds_evaluate
except ImportError:
print('Training package is not installed. See training documentation.')
raise

14
evaluate_tflite.py
View File

@ -10,23 +10,23 @@ import csv
import os
import sys
from deepspeech import Model
from deepspeech_training.util.evaluate_tools import calculate_and_print_report
from deepspeech_training.util.flags import create_flags
from stt import Model
from coqui_stt_training.util.evaluate_tools import calculate_and_print_report
from coqui_stt_training.util.flags import create_flags
from functools import partial
from multiprocessing import JoinableQueue, Process, cpu_count, Manager
from six.moves import zip, range
r'''
This module should be self-contained:
- build libdeepspeech.so with TFLite:
- bazel build [...] --define=runtime=tflite [...] //native_client:libdeepspeech.so
- build libstt.so with TFLite:
- bazel build [...] --define=runtime=tflite [...] //native_client:libstt.so
- make -C native_client/python/ TFDIR=... bindings
- setup a virtualenv
- pip install native_client/python/dist/deepspeech*.whl
- pip install native_client/python/dist/*.whl
- pip install -r requirements_eval_tflite.txt
Then run with a TF Lite model, a scorer and a CSV test file
Then run with a TFLite model, a scorer and a CSV test file
'''
def tflite_worker(model, scorer, queue_in, queue_out, gpu_mask):

4
examples/README.rst
View File

@ -1,6 +1,6 @@
Examples
========
DeepSpeech examples were moved to a separate repository.
🐸STT examples were moved to a separate repository.
New location: https://github.com/mozilla/DeepSpeech-examples
New location: https://github.com/coqui-ai/STT-examples

BIN
images/coqui-STT-logo-green.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 60 KiB

12
lm_optimizer.py
View File

@ -7,12 +7,12 @@ import optuna
import sys
import tensorflow.compat.v1 as tfv1
from deepspeech_training.evaluate import evaluate
from deepspeech_training.train import create_model
from deepspeech_training.util.config import Config, initialize_globals
from deepspeech_training.util.flags import create_flags, FLAGS
from deepspeech_training.util.logging import log_error
from deepspeech_training.util.evaluate_tools import wer_cer_batch
from coqui_stt_training.evaluate import evaluate
from coqui_stt_training.train import create_model
from coqui_stt_training.util.config import Config, initialize_globals
from coqui_stt_training.util.flags import create_flags, FLAGS
from coqui_stt_training.util.logging import log_error
from coqui_stt_training.util.evaluate_tools import wer_cer_batch
from ds_ctcdecoder import Scorer

8
native_client/Android.mk
View File

@ -1,14 +1,14 @@
LOCAL_PATH := $(call my-dir)
include $(CLEAR_VARS)
LOCAL_MODULE := deepspeech-prebuilt
LOCAL_SRC_FILES := $(TFDIR)/bazel-bin/native_client/libdeepspeech.so
LOCAL_MODULE := stt-prebuilt
LOCAL_SRC_FILES := $(TFDIR)/bazel-bin/native_client/libstt.so
include $(PREBUILT_SHARED_LIBRARY)
include $(CLEAR_VARS)
LOCAL_CPP_EXTENSION := .cc .cxx .cpp
LOCAL_MODULE := deepspeech
LOCAL_MODULE := stt
LOCAL_SRC_FILES := client.cc
LOCAL_SHARED_LIBRARIES := deepspeech-prebuilt
LOCAL_SHARED_LIBRARIES := stt-prebuilt
LOCAL_LDFLAGS := -Wl,--no-as-needed
include $(BUILD_EXECUTABLE)

30
native_client/BUILD
View File

@ -1,4 +1,4 @@
# Description: Deepspeech native client library.
# Description: Coqui STT native client library.
load("@org_tensorflow//tensorflow:tensorflow.bzl", "tf_cc_shared_object", "tf_copts", "lrt_if_needed")
load("@local_config_cuda//cuda:build_defs.bzl", "if_cuda")
@ -112,11 +112,11 @@ cc_library(
)
cc_library(
name = "deepspeech_bundle",
name = "coqui_stt_bundle",
srcs = [
"deepspeech.cc",
"deepspeech.h",
"deepspeech_errors.cc",
"stt.cc",
"coqui-stt.h",
"stt_errors.cc",
"modelstate.cc",
"modelstate.h",
"workspace_status.cc",
@ -165,7 +165,7 @@ cc_library(
#"//tensorflow/core:all_kernels",
### => Trying to be more fine-grained
### Use bin/ops_in_graph.py to list all the ops used by a frozen graph.
### CPU only build, libdeepspeech.so file size reduced by ~50%
### CPU only build, libstt.so file size reduced by ~50%
"//tensorflow/core/kernels:spectrogram_op", # AudioSpectrogram
"//tensorflow/core/kernels:bias_op", # BiasAdd
"//tensorflow/core/kernels:cast_op", # Cast
@ -205,31 +205,31 @@ cc_library(
)
tf_cc_shared_object(
name = "libdeepspeech.so",
deps = [":deepspeech_bundle"],
name = "libstt.so",
deps = [":coqui_stt_bundle"],
)
ios_static_framework(
name = "deepspeech_ios",
deps = [":deepspeech_bundle"],
name = "coqui_stt_ios",
deps = [":coqui_stt_bundle"],
families = ["iphone", "ipad"],
minimum_os_version = "9.0",
linkopts = ["-lstdc++"],
)
genrule(
name = "libdeepspeech_so_dsym",
srcs = [":libdeepspeech.so"],
outs = ["libdeepspeech.so.dSYM"],
name = "libstt_so_dsym",
srcs = [":libstt.so"],
outs = ["libstt.so.dSYM"],
output_to_bindir = True,
cmd = "dsymutil $(location :libdeepspeech.so) -o $@"
cmd = "dsymutil $(location :libstt.so) -o $@"
)
cc_binary(
name = "generate_scorer_package",
srcs = [
"generate_scorer_package.cpp",
"deepspeech_errors.cc",
"stt_errors.cc",
],
copts = ["-std=c++11"],
deps = [

4
native_client/CODINGSTYLE.md
View File

@ -1,5 +1,5 @@
This file contains some notes on coding style within the C++ portion of the
DeepSpeech project. It is very much a work in progress and incomplete.
🐸STT project. It is very much a work in progress and incomplete.
General
=======
@ -25,4 +25,4 @@ File naming
Doubts
======
If in doubt, please ask on our Matrix chat channel: https://chat.mozilla.org/#/room/#machinelearning:mozilla.org
If in doubt, please ask on our Matrix chat channel: https://matrix.to/#/#stt:matrix.org?via=matrix.org

30
native_client/Makefile
View File

@ -13,35 +13,35 @@
include definitions.mk
default: $(DEEPSPEECH_BIN)
default: $(STT_BIN)
clean:
rm -f deepspeech
rm -f stt
$(DEEPSPEECH_BIN): client.cc Makefile
$(CXX) $(CFLAGS) $(CFLAGS_DEEPSPEECH) $(SOX_CFLAGS) client.cc $(LDFLAGS) $(SOX_LDFLAGS)
$(STT_BIN): client.cc Makefile
$(CXX) $(CFLAGS) $(CFLAGS_STT) $(SOX_CFLAGS) client.cc $(LDFLAGS) $(SOX_LDFLAGS)
ifeq ($(OS),Darwin)
install_name_tool -change bazel-out/local-opt/bin/native_client/libdeepspeech.so @rpath/libdeepspeech.so deepspeech
install_name_tool -change bazel-out/local-opt/bin/native_client/libstt.so @rpath/libstt.so stt
endif
run: $(DEEPSPEECH_BIN)
${META_LD_LIBRARY_PATH}=${TFDIR}/bazel-bin/native_client:${${META_LD_LIBRARY_PATH}} ./deepspeech ${ARGS}
run: $(STT_BIN)
${META_LD_LIBRARY_PATH}=${TFDIR}/bazel-bin/native_client:${${META_LD_LIBRARY_PATH}} ./stt ${ARGS}
debug: $(DEEPSPEECH_BIN)
${META_LD_LIBRARY_PATH}=${TFDIR}/bazel-bin/native_client:${${META_LD_LIBRARY_PATH}} gdb --args ./deepspeech ${ARGS}
debug: $(STT_BIN)
${META_LD_LIBRARY_PATH}=${TFDIR}/bazel-bin/native_client:${${META_LD_LIBRARY_PATH}} gdb --args ./stt ${ARGS}
install: $(DEEPSPEECH_BIN)
install: $(STT_BIN)
install -d ${PREFIX}/lib
install -m 0644 ${TFDIR}/bazel-bin/native_client/libdeepspeech.so ${PREFIX}/lib/
install -m 0644 ${TFDIR}/bazel-bin/native_client/libstt.so ${PREFIX}/lib/
install -d ${PREFIX}/include
install -m 0644 deepspeech.h ${PREFIX}/include
install -m 0644 coqui-stt.h ${PREFIX}/include
install -d ${PREFIX}/bin
install -m 0755 deepspeech ${PREFIX}/bin/
install -m 0755 stt ${PREFIX}/bin/
uninstall:
rm -f ${PREFIX}/bin/deepspeech
rm -f ${PREFIX}/bin/stt
rmdir --ignore-fail-on-non-empty ${PREFIX}/bin
rm -f ${PREFIX}/lib/libdeepspeech.so
rm -f ${PREFIX}/lib/libstt.so
rmdir --ignore-fail-on-non-empty ${PREFIX}/lib
print-toolchain:

16
native_client/args.h
View File

@ -8,7 +8,7 @@
#endif
#include <iostream>
#include "deepspeech.h"
#include "coqui-stt.h"
char* model = NULL;
@ -47,7 +47,7 @@ void PrintHelp(const char* bin)
std::cout <<
"Usage: " << bin << " --model MODEL [--scorer SCORER] --audio AUDIO [-t] [-e]\n"
"\n"
"Running DeepSpeech inference.\n"
"Running Coqui STT inference.\n"
"\n"
"\t--model MODEL\t\t\tPath to the model (protocol buffer binary file)\n"
"\t--scorer SCORER\t\t\tPath to the external scorer file\n"
@ -64,9 +64,9 @@ void PrintHelp(const char* bin)
"\t--hot_words\t\t\tHot-words and their boosts. Word:Boost pairs are comma-separated\n"
"\t--help\t\t\t\tShow help\n"
"\t--version\t\t\tPrint version and exits\n";
char* version = DS_Version();
std::cerr << "DeepSpeech " << version << "\n";
DS_FreeString(version);
char* version = STT_Version();
std::cerr << "Coqui STT " << version << "\n";
STT_FreeString(version);
exit(1);
}
@ -169,9 +169,9 @@ bool ProcessArgs(int argc, char** argv)
}
if (has_versions) {
char* version = DS_Version();
std::cout << "DeepSpeech " << version << "\n";
DS_FreeString(version);
char* version = STT_Version();
std::cout << "Coqui " << version << "\n";
STT_FreeString(version);
return false;
}

4
native_client/bazel_workspace_status_cmd.sh
View File

@ -22,8 +22,8 @@ echo "STABLE_TF_GIT_VERSION ${tf_git_rev}"
pushd $(dirname "$0")
ds_git_rev=$(git describe --long --tags)
echo "STABLE_DS_GIT_VERSION ${ds_git_rev}"
ds_version=$(cat ../training/deepspeech_training/VERSION)
ds_version=$(cat ../training/coqui_stt_training/VERSION)
echo "STABLE_DS_VERSION ${ds_version}"
ds_graph_version=$(cat ../training/deepspeech_training/GRAPH_VERSION)
ds_graph_version=$(cat ../training/coqui_stt_training/GRAPH_VERSION)
echo "STABLE_DS_GRAPH_VERSION ${ds_graph_version}"
popd

64
native_client/client.cc
View File

@ -34,7 +34,7 @@
#endif // NO_DIR
#include <vector>
#include "deepspeech.h"
#include "coqui-stt.h"
#include "args.h"
typedef struct {
@ -168,17 +168,17 @@ LocalDsSTT(ModelState* aCtx, const short* aBuffer, size_t aBufferSize,
// sphinx-doc: c_ref_inference_start
if (extended_output) {
Metadata *result = DS_SpeechToTextWithMetadata(aCtx, aBuffer, aBufferSize, 1);
Metadata *result = STT_SpeechToTextWithMetadata(aCtx, aBuffer, aBufferSize, 1);
res.string = CandidateTranscriptToString(&result->transcripts[0]);
DS_FreeMetadata(result);
STT_FreeMetadata(result);
} else if (json_output) {
Metadata *result = DS_SpeechToTextWithMetadata(aCtx, aBuffer, aBufferSize, json_candidate_transcripts);
Metadata *result = STT_SpeechToTextWithMetadata(aCtx, aBuffer, aBufferSize, json_candidate_transcripts);
res.string = MetadataToJSON(result);
DS_FreeMetadata(result);
STT_FreeMetadata(result);
} else if (stream_size > 0) {
StreamingState* ctx;
int status = DS_CreateStream(aCtx, &ctx);
if (status != DS_ERR_OK) {
int status = STT_CreateStream(aCtx, &ctx);
if (status != STT_ERR_OK) {
res.string = strdup("");
return res;
}
@ -187,28 +187,28 @@ LocalDsSTT(ModelState* aCtx, const short* aBuffer, size_t aBufferSize,
const char *prev = nullptr;
while (off < aBufferSize) {
size_t cur = aBufferSize - off > stream_size ? stream_size : aBufferSize - off;
DS_FeedAudioContent(ctx, aBuffer + off, cur);
STT_FeedAudioContent(ctx, aBuffer + off, cur);
off += cur;
prev = last;
const char* partial = DS_IntermediateDecode(ctx);
const char* partial = STT_IntermediateDecode(ctx);
if (last == nullptr || strcmp(last, partial)) {
printf("%s\n", partial);
last = partial;
} else {
DS_FreeString((char *) partial);
STT_FreeString((char *) partial);
}
if (prev != nullptr && prev != last) {
DS_FreeString((char *) prev);
STT_FreeString((char *) prev);
}
}
if (last != nullptr) {
DS_FreeString((char *) last);
STT_FreeString((char *) last);
}
res.string = DS_FinishStream(ctx);
res.string = STT_FinishStream(ctx);
} else if (extended_stream_size > 0) {
StreamingState* ctx;
int status = DS_CreateStream(aCtx, &ctx);
if (status != DS_ERR_OK) {
int status = STT_CreateStream(aCtx, &ctx);
if (status != STT_ERR_OK) {
res.string = strdup("");
return res;
}
@ -217,10 +217,10 @@ LocalDsSTT(ModelState* aCtx, const short* aBuffer, size_t aBufferSize,
const char *prev = nullptr;
while (off < aBufferSize) {
size_t cur = aBufferSize - off > extended_stream_size ? extended_stream_size : aBufferSize - off;
DS_FeedAudioContent(ctx, aBuffer + off, cur);
STT_FeedAudioContent(ctx, aBuffer + off, cur);
off += cur;
prev = last;
const Metadata* result = DS_IntermediateDecodeWithMetadata(ctx, 1);
const Metadata* result = STT_IntermediateDecodeWithMetadata(ctx, 1);
const char* partial = CandidateTranscriptToString(&result->transcripts[0]);
if (last == nullptr || strcmp(last, partial)) {
printf("%s\n", partial);
@ -231,14 +231,14 @@ LocalDsSTT(ModelState* aCtx, const short* aBuffer, size_t aBufferSize,
if (prev != nullptr && prev != last) {
free((char *) prev);
}
DS_FreeMetadata((Metadata *)result);
STT_FreeMetadata((Metadata *)result);
}
const Metadata* result = DS_FinishStreamWithMetadata(ctx, 1);
const Metadata* result = STT_FinishStreamWithMetadata(ctx, 1);
res.string = CandidateTranscriptToString(&result->transcripts[0]);
DS_FreeMetadata((Metadata *)result);
STT_FreeMetadata((Metadata *)result);
free((char *) last);
} else {
res.string = DS_SpeechToText(aCtx, aBuffer, aBufferSize);
res.string = STT_SpeechToText(aCtx, aBuffer, aBufferSize);
}
// sphinx-doc: c_ref_inference_stop
@ -404,9 +404,9 @@ GetAudioBuffer(const char* path, int desired_sample_rate)
void
ProcessFile(ModelState* context, const char* path, bool show_times)
{
ds_audio_buffer audio = GetAudioBuffer(path, DS_GetModelSampleRate(context));
ds_audio_buffer audio = GetAudioBuffer(path, STT_GetModelSampleRate(context));
// Pass audio to DeepSpeech
// Pass audio to STT
// We take half of buffer_size because buffer is a char* while
// LocalDsSTT() expected a short*
ds_result result = LocalDsSTT(context,
@ -418,7 +418,7 @@ ProcessFile(ModelState* context, const char* path, bool show_times)
if (result.string) {
printf("%s\n", result.string);
DS_FreeString((char*)result.string);
STT_FreeString((char*)result.string);
}
if (show_times) {
@ -450,19 +450,19 @@ main(int argc, char **argv)
return 1;
}
// Initialise DeepSpeech
// Initialise STT
ModelState* ctx;
// sphinx-doc: c_ref_model_start
int status = DS_CreateModel(model, &ctx);
int status = STT_CreateModel(model, &ctx);
if (status != 0) {
char* error = DS_ErrorCodeToErrorMessage(status);
char* error = STT_ErrorCodeToErrorMessage(status);
fprintf(stderr, "Could not create model: %s\n", error);
free(error);
return 1;
}
if (set_beamwidth) {
status = DS_SetModelBeamWidth(ctx, beam_width);
status = STT_SetModelBeamWidth(ctx, beam_width);
if (status != 0) {
fprintf(stderr, "Could not set model beam width.\n");
return 1;
@ -470,13 +470,13 @@ main(int argc, char **argv)
}
if (scorer) {
status = DS_EnableExternalScorer(ctx, scorer);
status = STT_EnableExternalScorer(ctx, scorer);
if (status != 0) {
fprintf(stderr, "Could not enable external scorer.\n");
return 1;
}
if (set_alphabeta) {
status = DS_SetScorerAlphaBeta(ctx, lm_alpha, lm_beta);
status = STT_SetScorerAlphaBeta(ctx, lm_alpha, lm_beta);
if (status != 0) {
fprintf(stderr, "Error setting scorer alpha and beta.\n");
return 1;
@ -494,7 +494,7 @@ main(int argc, char **argv)
// so, check the boost string before we turn it into a float
bool boost_is_valid = (pair_[1].find_first_not_of("-.0123456789") == std::string::npos);
float boost = strtof((pair_[1]).c_str(),0);
status = DS_AddHotWord(ctx, word, boost);
status = STT_AddHotWord(ctx, word, boost);
if (status != 0 || !boost_is_valid) {
fprintf(stderr, "Could not enable hot-word.\n");
return 1;
@ -555,7 +555,7 @@ main(int argc, char **argv)
sox_quit();
#endif // NO_SOX
DS_FreeModel(ctx);
STT_FreeModel(ctx);
return 0;
}

210
native_client/deepspeech.h → native_client/coqui-stt.h
View File

@ -1,5 +1,5 @@
#ifndef DEEPSPEECH_H
#define DEEPSPEECH_H
#ifndef COQUI_STT_H
#define COQUI_STT_H
#ifdef __cplusplus
extern "C" {
@ -7,12 +7,12 @@ extern "C" {
#ifndef SWIG
#if defined _MSC_VER
#define DEEPSPEECH_EXPORT __declspec(dllexport)
#define STT_EXPORT __declspec(dllexport)
#else
#define DEEPSPEECH_EXPORT __attribute__ ((visibility("default")))
#define STT_EXPORT __attribute__ ((visibility("default")))
#endif /*End of _MSC_VER*/
#else
#define DEEPSPEECH_EXPORT
#define STT_EXPORT
#endif
typedef struct ModelState ModelState;
@ -61,92 +61,92 @@ typedef struct Metadata {
// sphinx-doc: error_code_listing_start
#define DS_FOR_EACH_ERROR(APPLY) \
APPLY(DS_ERR_OK, 0x0000, "No error.") \
APPLY(DS_ERR_NO_MODEL, 0x1000, "Missing model information.") \
APPLY(DS_ERR_INVALID_ALPHABET, 0x2000, "Invalid alphabet embedded in model. (Data corruption?)") \
APPLY(DS_ERR_INVALID_SHAPE, 0x2001, "Invalid model shape.") \
APPLY(DS_ERR_INVALID_SCORER, 0x2002, "Invalid scorer file.") \
APPLY(DS_ERR_MODEL_INCOMPATIBLE, 0x2003, "Incompatible model.") \
APPLY(DS_ERR_SCORER_NOT_ENABLED, 0x2004, "External scorer is not enabled.") \
APPLY(DS_ERR_SCORER_UNREADABLE, 0x2005, "Could not read scorer file.") \
APPLY(DS_ERR_SCORER_INVALID_LM, 0x2006, "Could not recognize language model header in scorer.") \
APPLY(DS_ERR_SCORER_NO_TRIE, 0x2007, "Reached end of scorer file before loading vocabulary trie.") \
APPLY(DS_ERR_SCORER_INVALID_TRIE, 0x2008, "Invalid magic in trie header.") \
APPLY(DS_ERR_SCORER_VERSION_MISMATCH, 0x2009, "Scorer file version does not match expected version.") \
APPLY(DS_ERR_FAIL_INIT_MMAP, 0x3000, "Failed to initialize memory mapped model.") \
APPLY(DS_ERR_FAIL_INIT_SESS, 0x3001, "Failed to initialize the session.") \
APPLY(DS_ERR_FAIL_INTERPRETER, 0x3002, "Interpreter failed.") \
APPLY(DS_ERR_FAIL_RUN_SESS, 0x3003, "Failed to run the session.") \
APPLY(DS_ERR_FAIL_CREATE_STREAM, 0x3004, "Error creating the stream.") \
APPLY(DS_ERR_FAIL_READ_PROTOBUF, 0x3005, "Error reading the proto buffer model file.") \
APPLY(DS_ERR_FAIL_CREATE_SESS, 0x3006, "Failed to create session.") \
APPLY(DS_ERR_FAIL_CREATE_MODEL, 0x3007, "Could not allocate model state.") \
APPLY(DS_ERR_FAIL_INSERT_HOTWORD, 0x3008, "Could not insert hot-word.") \
APPLY(DS_ERR_FAIL_CLEAR_HOTWORD, 0x3009, "Could not clear hot-words.") \
APPLY(DS_ERR_FAIL_ERASE_HOTWORD, 0x3010, "Could not erase hot-word.")
#define STT_FOR_EACH_ERROR(APPLY) \
APPLY(STT_ERR_OK, 0x0000, "No error.") \
APPLY(STT_ERR_NO_MODEL, 0x1000, "Missing model information.") \
APPLY(STT_ERR_INVALID_ALPHABET, 0x2000, "Invalid alphabet embedded in model. (Data corruption?)") \
APPLY(STT_ERR_INVALID_SHAPE, 0x2001, "Invalid model shape.") \
APPLY(STT_ERR_INVALID_SCORER, 0x2002, "Invalid scorer file.") \
APPLY(STT_ERR_MODEL_INCOMPATIBLE, 0x2003, "Incompatible model.") \
APPLY(STT_ERR_SCORER_NOT_ENABLED, 0x2004, "External scorer is not enabled.") \
APPLY(STT_ERR_SCORER_UNREADABLE, 0x2005, "Could not read scorer file.") \
APPLY(STT_ERR_SCORER_INVALID_LM, 0x2006, "Could not recognize language model header in scorer.") \
APPLY(STT_ERR_SCORER_NO_TRIE, 0x2007, "Reached end of scorer file before loading vocabulary trie.") \
APPLY(STT_ERR_SCORER_INVALID_TRIE, 0x2008, "Invalid magic in trie header.") \
APPLY(STT_ERR_SCORER_VERSION_MISMATCH, 0x2009, "Scorer file version does not match expected version.") \
APPLY(STT_ERR_FAIL_INIT_MMAP, 0x3000, "Failed to initialize memory mapped model.") \
APPLY(STT_ERR_FAIL_INIT_SESS, 0x3001, "Failed to initialize the session.") \
APPLY(STT_ERR_FAIL_INTERPRETER, 0x3002, "Interpreter failed.") \
APPLY(STT_ERR_FAIL_RUN_SESS, 0x3003, "Failed to run the session.") \
APPLY(STT_ERR_FAIL_CREATE_STREAM, 0x3004, "Error creating the stream.") \
APPLY(STT_ERR_FAIL_READ_PROTOBUF, 0x3005, "Error reading the proto buffer model file.") \
APPLY(STT_ERR_FAIL_CREATE_SESS, 0x3006, "Failed to create session.") \
APPLY(STT_ERR_FAIL_CREATE_MODEL, 0x3007, "Could not allocate model state.") \
APPLY(STT_ERR_FAIL_INSERT_HOTWORD, 0x3008, "Could not insert hot-word.") \
APPLY(STT_ERR_FAIL_CLEAR_HOTWORD, 0x3009, "Could not clear hot-words.") \
APPLY(STT_ERR_FAIL_ERASE_HOTWORD, 0x3010, "Could not erase hot-word.")
// sphinx-doc: error_code_listing_end
enum DeepSpeech_Error_Codes
enum STT_Error_Codes
{
#define DEFINE(NAME, VALUE, DESC) NAME = VALUE,
DS_FOR_EACH_ERROR(DEFINE)
STT_FOR_EACH_ERROR(DEFINE)
#undef DEFINE
};
/**
* @brief An object providing an interface to a trained DeepSpeech model.
* @brief An object providing an interface to a trained Coqui STT model.
*
* @param aModelPath The path to the frozen model graph.
* @param[out] retval a ModelState pointer
*
* @return Zero on success, non-zero on failure.
*/
DEEPSPEECH_EXPORT
int DS_CreateModel(const char* aModelPath,
STT_EXPORT
int STT_CreateModel(const char* aModelPath,
ModelState** retval);
/**
* @brief Get beam width value used by the model. If {@link DS_SetModelBeamWidth}
* @brief Get beam width value used by the model. If {@link STT_SetModelBeamWidth}
* was not called before, will return the default value loaded from the
* model file.
*
* @param aCtx A ModelState pointer created with {@link DS_CreateModel}.
* @param aCtx A ModelState pointer created with {@link STT_CreateModel}.
*
* @return Beam width value used by the model.
*/
DEEPSPEECH_EXPORT
unsigned int DS_GetModelBeamWidth(const ModelState* aCtx);
STT_EXPORT
unsigned int STT_GetModelBeamWidth(const ModelState* aCtx);
/**
* @brief Set beam width value used by the model.
*
* @param aCtx A ModelState pointer created with {@link DS_CreateModel}.
* @param aCtx A ModelState pointer created with {@link STT_CreateModel}.
* @param aBeamWidth The beam width used by the model. A larger beam width value
* generates better results at the cost of decoding time.
*
* @return Zero on success, non-zero on failure.
*/
DEEPSPEECH_EXPORT
int DS_SetModelBeamWidth(ModelState* aCtx,
STT_EXPORT
int STT_SetModelBeamWidth(ModelState* aCtx,
unsigned int aBeamWidth);
/**
* @brief Return the sample rate expected by a model.
*
* @param aCtx A ModelState pointer created with {@link DS_CreateModel}.
* @param aCtx A ModelState pointer created with {@link STT_CreateModel}.
*
* @return Sample rate expected by the model for its input.
*/
DEEPSPEECH_EXPORT
int DS_GetModelSampleRate(const ModelState* aCtx);
STT_EXPORT
int STT_GetModelSampleRate(const ModelState* aCtx);
/**
* @brief Frees associated resources and destroys model object.
*/
DEEPSPEECH_EXPORT
void DS_FreeModel(ModelState* ctx);
STT_EXPORT
void STT_FreeModel(ModelState* ctx);
/**
* @brief Enable decoding using an external scorer.
@ -156,8 +156,8 @@ void DS_FreeModel(ModelState* ctx);
*
* @return Zero on success, non-zero on failure (invalid arguments).
*/
DEEPSPEECH_EXPORT
int DS_EnableExternalScorer(ModelState* aCtx,
STT_EXPORT
int STT_EnableExternalScorer(ModelState* aCtx,
const char* aScorerPath);
/**
@ -171,8 +171,8 @@ int DS_EnableExternalScorer(ModelState* aCtx,
*
* @return Zero on success, non-zero on failure (invalid arguments).
*/
DEEPSPEECH_EXPORT
int DS_AddHotWord(ModelState* aCtx,
STT_EXPORT
int STT_AddHotWord(ModelState* aCtx,
const char* word,
float boost);
@ -184,8 +184,8 @@ int DS_AddHotWord(ModelState* aCtx,
*
* @return Zero on success, non-zero on failure (invalid arguments).
*/
DEEPSPEECH_EXPORT
int DS_EraseHotWord(ModelState* aCtx,
STT_EXPORT
int STT_EraseHotWord(ModelState* aCtx,
const char* word);
/**
@ -195,8 +195,8 @@ int DS_EraseHotWord(ModelState* aCtx,
*
* @return Zero on success, non-zero on failure (invalid arguments).
*/
DEEPSPEECH_EXPORT
int DS_ClearHotWords(ModelState* aCtx);
STT_EXPORT
int STT_ClearHotWords(ModelState* aCtx);
/**
* @brief Disable decoding using an external scorer.
@ -205,8 +205,8 @@ int DS_ClearHotWords(ModelState* aCtx);
*
* @return Zero on success, non-zero on failure.
*/
DEEPSPEECH_EXPORT
int DS_DisableExternalScorer(ModelState* aCtx);
STT_EXPORT
int STT_DisableExternalScorer(ModelState* aCtx);
/**
* @brief Set hyperparameters alpha and beta of the external scorer.
@ -217,13 +217,13 @@ int DS_DisableExternalScorer(ModelState* aCtx);
*
* @return Zero on success, non-zero on failure.
*/
DEEPSPEECH_EXPORT
int DS_SetScorerAlphaBeta(ModelState* aCtx,
STT_EXPORT
int STT_SetScorerAlphaBeta(ModelState* aCtx,
float aAlpha,
float aBeta);
/**
* @brief Use the DeepSpeech model to convert speech to text.
* @brief Use the Coqui STT model to convert speech to text.
*
* @param aCtx The ModelState pointer for the model to use.
* @param aBuffer A 16-bit, mono raw audio signal at the appropriate
@ -231,15 +231,15 @@ int DS_SetScorerAlphaBeta(ModelState* aCtx,
* @param aBufferSize The number of samples in the audio signal.
*
* @return The STT result. The user is responsible for freeing the string using
* {@link DS_FreeString()}. Returns NULL on error.
* {@link STT_FreeString()}. Returns NULL on error.
*/
DEEPSPEECH_EXPORT
char* DS_SpeechToText(ModelState* aCtx,
STT_EXPORT
char* STT_SpeechToText(ModelState* aCtx,
const short* aBuffer,
unsigned int aBufferSize);
/**
* @brief Use the DeepSpeech model to convert speech to text and output results
* @brief Use the Coqui STT model to convert speech to text and output results
* including metadata.
*
* @param aCtx The ModelState pointer for the model to use.
@ -250,19 +250,19 @@ char* DS_SpeechToText(ModelState* aCtx,
*
* @return Metadata struct containing multiple CandidateTranscript structs. Each
* transcript has per-token metadata including timing information. The
* user is responsible for freeing Metadata by calling {@link DS_FreeMetadata()}.
* user is responsible for freeing Metadata by calling {@link STT_FreeMetadata()}.
* Returns NULL on error.
*/
DEEPSPEECH_EXPORT
Metadata* DS_SpeechToTextWithMetadata(ModelState* aCtx,
STT_EXPORT
Metadata* STT_SpeechToTextWithMetadata(ModelState* aCtx,
const short* aBuffer,
unsigned int aBufferSize,
unsigned int aNumResults);
/**
* @brief Create a new streaming inference state. The streaming state returned
* by this function can then be passed to {@link DS_FeedAudioContent()}
* and {@link DS_FinishStream()}.
* by this function can then be passed to {@link STT_FeedAudioContent()}
* and {@link STT_FinishStream()}.
*
* @param aCtx The ModelState pointer for the model to use.
* @param[out] retval an opaque pointer that represents the streaming state. Can
@ -270,81 +270,81 @@ Metadata* DS_SpeechToTextWithMetadata(ModelState* aCtx,
*
* @return Zero for success, non-zero on failure.
*/
DEEPSPEECH_EXPORT
int DS_CreateStream(ModelState* aCtx,
STT_EXPORT
int STT_CreateStream(ModelState* aCtx,
StreamingState** retval);
/**
* @brief Feed audio samples to an ongoing streaming inference.
*
* @param aSctx A streaming state pointer returned by {@link DS_CreateStream()}.
* @param aSctx A streaming state pointer returned by {@link STT_CreateStream()}.
* @param aBuffer An array of 16-bit, mono raw audio samples at the
* appropriate sample rate (matching what the model was trained on).
* @param aBufferSize The number of samples in @p aBuffer.
*/
DEEPSPEECH_EXPORT
void DS_FeedAudioContent(StreamingState* aSctx,
STT_EXPORT
void STT_FeedAudioContent(StreamingState* aSctx,
const short* aBuffer,
unsigned int aBufferSize);
/**
* @brief Compute the intermediate decoding of an ongoing streaming inference.
*
* @param aSctx A streaming state pointer returned by {@link DS_CreateStream()}.
* @param aSctx A streaming state pointer returned by {@link STT_CreateStream()}.
*
* @return The STT intermediate result. The user is responsible for freeing the
* string using {@link DS_FreeString()}.
* string using {@link STT_FreeString()}.
*/
DEEPSPEECH_EXPORT
char* DS_IntermediateDecode(const StreamingState* aSctx);
STT_EXPORT
char* STT_IntermediateDecode(const StreamingState* aSctx);
/**
* @brief Compute the intermediate decoding of an ongoing streaming inference,
* return results including metadata.
*
* @param aSctx A streaming state pointer returned by {@link DS_CreateStream()}.
* @param aSctx A streaming state pointer returned by {@link STT_CreateStream()}.
* @param aNumResults The number of candidate transcripts to return.
*
* @return Metadata struct containing multiple candidate transcripts. Each transcript
* has per-token metadata including timing information. The user is
* responsible for freeing Metadata by calling {@link DS_FreeMetadata()}.
* responsible for freeing Metadata by calling {@link STT_FreeMetadata()}.
* Returns NULL on error.
*/
DEEPSPEECH_EXPORT
Metadata* DS_IntermediateDecodeWithMetadata(const StreamingState* aSctx,
STT_EXPORT
Metadata* STT_IntermediateDecodeWithMetadata(const StreamingState* aSctx,
unsigned int aNumResults);
/**
* @brief Compute the final decoding of an ongoing streaming inference and return
* the result. Signals the end of an ongoing streaming inference.
*
* @param aSctx A streaming state pointer returned by {@link DS_CreateStream()}.
* @param aSctx A streaming state pointer returned by {@link STT_CreateStream()}.
*
* @return The STT result. The user is responsible for freeing the string using
* {@link DS_FreeString()}.
* {@link STT_FreeString()}.
*
* @note This method will free the state pointer (@p aSctx).
*/
DEEPSPEECH_EXPORT
char* DS_FinishStream(StreamingState* aSctx);
STT_EXPORT
char* STT_FinishStream(StreamingState* aSctx);
/**
* @brief Compute the final decoding of an ongoing streaming inference and return
* results including metadata. Signals the end of an ongoing streaming
* inference.
*
* @param aSctx A streaming state pointer returned by {@link DS_CreateStream()}.
* @param aSctx A streaming state pointer returned by {@link STT_CreateStream()}.
* @param aNumResults The number of candidate transcripts to return.
*
* @return Metadata struct containing multiple candidate transcripts. Each transcript
* has per-token metadata including timing information. The user is
* responsible for freeing Metadata by calling {@link DS_FreeMetadata()}.
* responsible for freeing Metadata by calling {@link STT_FreeMetadata()}.
* Returns NULL on error.
*
* @note This method will free the state pointer (@p aSctx).
*/
DEEPSPEECH_EXPORT
Metadata* DS_FinishStreamWithMetadata(StreamingState* aSctx,
STT_EXPORT
Metadata* STT_FinishStreamWithMetadata(StreamingState* aSctx,
unsigned int aNumResults);
/**
@ -352,47 +352,47 @@ Metadata* DS_FinishStreamWithMetadata(StreamingState* aSctx,
* can be used if you no longer need the result of an ongoing streaming
* inference and don't want to perform a costly decode operation.
*
* @param aSctx A streaming state pointer returned by {@link DS_CreateStream()}.
* @param aSctx A streaming state pointer returned by {@link STT_CreateStream()}.
*
* @note This method will free the state pointer (@p aSctx).
*/
DEEPSPEECH_EXPORT
void DS_FreeStream(StreamingState* aSctx);
STT_EXPORT
void STT_FreeStream(StreamingState* aSctx);
/**
* @brief Free memory allocated for metadata information.
*/
DEEPSPEECH_EXPORT
void DS_FreeMetadata(Metadata* m);
STT_EXPORT
void STT_FreeMetadata(Metadata* m);
/**
* @brief Free a char* string returned by the DeepSpeech API.
* @brief Free a char* string returned by the Coqui STT API.
*/
DEEPSPEECH_EXPORT
void DS_FreeString(char* str);
STT_EXPORT
void STT_FreeString(char* str);
/**
* @brief Returns the version of this library. The returned version is a semantic
* version (SemVer 2.0.0). The string returned must be freed with {@link DS_FreeString()}.
* version (SemVer 2.0.0). The string returned must be freed with {@link STT_FreeString()}.
*
* @return The version string.
*/
DEEPSPEECH_EXPORT
char* DS_Version();
STT_EXPORT
char* STT_Version();
/**
* @brief Returns a textual description corresponding to an error code.
* The string returned must be freed with @{link DS_FreeString()}.
* The string returned must be freed with @{link STT_FreeString()}.
*
* @return The error description.
*/
DEEPSPEECH_EXPORT
char* DS_ErrorCodeToErrorMessage(int aErrorCode);
STT_EXPORT
char* STT_ErrorCodeToErrorMessage(int aErrorCode);
#undef DEEPSPEECH_EXPORT
#undef STT_EXPORT
#ifdef __cplusplus
}
#endif
#endif /* DEEPSPEECH_H */
#endif /* COQUI_STT_H */

2
native_client/ctcdecode/__init__.py
View File

@ -9,7 +9,7 @@ __version__ = swigwrapper.__version__.decode('utf-8')
# Hack: import error codes by matching on their names, as SWIG unfortunately
# does not support binding enums to Python in a scoped manner yet.
for symbol in dir(swigwrapper):
if symbol.startswith('DS_ERR_'):
if symbol.startswith('STT_ERR_'):
globals()[symbol] = getattr(swigwrapper, symbol)
class Scorer(swigwrapper.Scorer):

14
native_client/ctcdecode/scorer.cpp
View File

@ -74,13 +74,13 @@ int Scorer::load_lm(const std::string& lm_path)
// Check if file is readable to avoid KenLM throwing an exception
const char* filename = lm_path.c_str();
if (access(filename, R_OK) != 0) {
return DS_ERR_SCORER_UNREADABLE;
return STT_ERR_SCORER_UNREADABLE;
}
// Check if the file format is valid to avoid KenLM throwing an exception
lm::ngram::ModelType model_type;
if (!lm::ngram::RecognizeBinary(filename, model_type)) {
return DS_ERR_SCORER_INVALID_LM;
return STT_ERR_SCORER_INVALID_LM;
}
// Load the LM
@ -97,7 +97,7 @@ int Scorer::load_lm(const std::string& lm_path)
uint64_t trie_offset = language_model_->GetEndOfSearchOffset();
if (package_size <= trie_offset) {
// File ends without a trie structure
return DS_ERR_SCORER_NO_TRIE;
return STT_ERR_SCORER_NO_TRIE;
}
// Read metadata and trie from file
@ -113,7 +113,7 @@ int Scorer::load_trie(std::ifstream& fin, const std::string& file_path)
if (magic != MAGIC) {
std::cerr << "Error: Can't parse scorer file, invalid header. Try updating "
"your scorer file." << std::endl;
return DS_ERR_SCORER_INVALID_TRIE;
return STT_ERR_SCORER_INVALID_TRIE;
}
int version;
@ -125,10 +125,10 @@ int Scorer::load_trie(std::ifstream& fin, const std::string& file_path)
if (version < FILE_VERSION) {
std::cerr << "Update your scorer file.";
} else {
std::cerr << "Downgrade your scorer file or update your version of DeepSpeech.";
std::cerr << "Downgrade your scorer file or update your version of Coqui STT.";
}
std::cerr << std::endl;
return DS_ERR_SCORER_VERSION_MISMATCH;
return STT_ERR_SCORER_VERSION_MISMATCH;
}
fin.read(reinterpret_cast<char*>(&is_utf8_mode_), sizeof(is_utf8_mode_));
@ -143,7 +143,7 @@ int Scorer::load_trie(std::ifstream& fin, const std::string& file_path)
opt.mode = fst::FstReadOptions::MAP;
opt.source = file_path;
dictionary.reset(FstType::Read(fin, opt));
return DS_ERR_OK;
return STT_ERR_OK;
}
bool Scorer::save_dictionary(const std::string& path, bool append_instead_of_overwrite)

2
native_client/ctcdecode/scorer.h
View File

@ -13,7 +13,7 @@
#include "path_trie.h"
#include "alphabet.h"
#include "deepspeech.h"
#include "coqui-stt.h"
const double OOV_SCORE = -1000.0;
const std::string START_TOKEN = "<s>";

Some files were not shown because too many files have changed in this diff Show More