Merge pull request #2354 from lissyx/run-examples-taskcluster

Run examples on TaskCluster
This commit is contained in:
lissyx 2019-09-18 20:56:22 +02:00 committed by GitHub
commit f98bfefc77
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
25 changed files with 325 additions and 53 deletions

View File

@ -17,16 +17,6 @@ const LM_ALPHA = 0.75;
// The beta hyperparameter of the CTC decoder. Word insertion bonus.
const LM_BETA = 1.85;
// These constants are tied to the shape of the graph used (changing them changes
// the geometry of the first layer), so make sure you use the same constants that
// were used during training
// Number of MFCC features to use
const N_FEATURES = 26;
// Size of the context window used for producing timesteps in the input vector
const N_CONTEXT = 9;
let VersionAction = function VersionAction(options) {
options = options || {};
options.nargs = 0;
@ -55,15 +45,14 @@ function totalTime(hrtimeValue) {
console.error('Loading model from file %s', args['model']);
const model_load_start = process.hrtime();
let model = new Ds.Model(args['model'], N_FEATURES, N_CONTEXT, args['alphabet'], BEAM_WIDTH);
let model = new Ds.Model(args['model'], args['alphabet'], BEAM_WIDTH);
const model_load_end = process.hrtime(model_load_start);
console.error('Loaded model in %ds.', totalTime(model_load_end));
if (args['lm'] && args['trie']) {
console.error('Loading language model from files %s %s', args['lm'], args['trie']);
const lm_load_start = process.hrtime();
model.enableDecoderWithLM(args['alphabet'], args['lm'], args['trie'],
LM_ALPHA, LM_BETA);
model.enableDecoderWithLM(args['lm'], args['trie'], LM_ALPHA, LM_BETA);
const lm_load_end = process.hrtime(lm_load_start);
console.error('Loaded language model in %ds.', totalTime(lm_load_end));
}
@ -106,7 +95,7 @@ const ffmpeg = spawn('ffmpeg', [
]);
let audioLength = 0;
let sctx = model.setupStream(AUDIO_SAMPLE_RATE);
let sctx = model.createStream(AUDIO_SAMPLE_RATE);
function finishStream() {
const model_load_start = process.hrtime();
@ -119,7 +108,7 @@ function finishStream() {
function intermediateDecode() {
finishStream();
sctx = model.setupStream(AUDIO_SAMPLE_RATE);
sctx = model.createStream(AUDIO_SAMPLE_RATE);
}
function feedAudioContent(chunk) {

View File

@ -8,7 +8,7 @@
},
"dependencies": {
"argparse": "^1.0.10",
"deepspeech": "^0.4.1",
"deepspeech": "^0.6.0-alpha.5",
"node-vad": "^1.1.1",
"util": "^0.11.1"
},

View File

@ -0,0 +1,30 @@
#!/bin/bash
set -xe
THIS=$(dirname "$0")
pushd ${THIS}
source ../tests.sh
npm install $(get_npm_package_url)
npm install
node ./index.js --audio $HOME/DeepSpeech/audio/2830-3980-0043.wav \
--lm $HOME/DeepSpeech/models/lm.binary \
--trie $HOME/DeepSpeech/models/trie \
--model $HOME/DeepSpeech/models/output_graph.pbmm \
--alphabet $HOME/DeepSpeech/models/alphabet.txt
node ./index.js --audio $HOME/DeepSpeech/audio/4507-16021-0012.wav \
--lm $HOME/DeepSpeech/models/lm.binary \
--trie $HOME/DeepSpeech/models/trie \
--model $HOME/DeepSpeech/models/output_graph.pbmm \
--alphabet $HOME/DeepSpeech/models/alphabet.txt
node ./index.js --audio $HOME/DeepSpeech/audio/8455-210777-0068.wav \
--lm $HOME/DeepSpeech/models/lm.binary \
--trie $HOME/DeepSpeech/models/trie \
--model $HOME/DeepSpeech/models/output_graph.pbmm \
--alphabet $HOME/DeepSpeech/models/alphabet.txt
popd

View File

@ -20,8 +20,11 @@ class Audio(object):
CHANNELS = 1
BLOCKS_PER_SECOND = 50
def __init__(self, callback=None, device=None, input_rate=RATE_PROCESS):
def __init__(self, callback=None, device=None, input_rate=RATE_PROCESS, file=None):
def proxy_callback(in_data, frame_count, time_info, status):
#pylint: disable=unused-argument
if self.chunk is not None:
in_data = self.wf.readframes(self.chunk)
callback(in_data)
return (None, pyaudio.paContinue)
if callback is None: callback = lambda in_data: self.buffer_queue.put(in_data)
@ -42,9 +45,13 @@ class Audio(object):
'stream_callback': proxy_callback,
}
self.chunk = None
# if not default device
if self.device:
kwargs['input_device_index'] = self.device
elif file is not None:
self.chunk = 320
self.wf = wave.open(file, 'rb')
self.stream = self.pa.open(**kwargs)
self.stream.start_stream()
@ -96,8 +103,8 @@ class Audio(object):
class VADAudio(Audio):
"""Filter & segment audio with voice activity detection."""
def __init__(self, aggressiveness=3, device=None, input_rate=None):
super().__init__(device=device, input_rate=input_rate)
def __init__(self, aggressiveness=3, device=None, input_rate=None, file=None):
super().__init__(device=device, input_rate=input_rate, file=file)
self.vad = webrtcvad.Vad(aggressiveness)
def frame_generator(self):
@ -121,6 +128,9 @@ class VADAudio(Audio):
triggered = False
for frame in frames:
if len(frame) < 640:
return
is_speech = self.vad.is_speech(frame, self.sample_rate)
if not triggered:
@ -153,23 +163,25 @@ def main(ARGS):
print('Initializing model...')
logging.info("ARGS.model: %s", ARGS.model)
logging.info("ARGS.alphabet: %s", ARGS.alphabet)
model = deepspeech.Model(ARGS.model, ARGS.n_features, ARGS.n_context, ARGS.alphabet, ARGS.beam_width)
model = deepspeech.Model(ARGS.model, ARGS.alphabet, ARGS.beam_width)
if ARGS.lm and ARGS.trie:
logging.info("ARGS.lm: %s", ARGS.lm)
logging.info("ARGS.trie: %s", ARGS.trie)
model.enableDecoderWithLM(ARGS.alphabet, ARGS.lm, ARGS.trie, ARGS.lm_alpha, ARGS.lm_beta)
model.enableDecoderWithLM(ARGS.lm, ARGS.trie, ARGS.lm_alpha, ARGS.lm_beta)
# Start audio with VAD
vad_audio = VADAudio(aggressiveness=ARGS.vad_aggressiveness,
device=ARGS.device,
input_rate=ARGS.rate)
input_rate=ARGS.rate,
file=ARGS.file)
print("Listening (ctrl-C to exit)...")
frames = vad_audio.vad_collector()
# Stream from microphone to DeepSpeech using VAD
spinner = None
if not ARGS.nospinner: spinner = Halo(spinner='line')
stream_context = model.setupStream()
if not ARGS.nospinner:
spinner = Halo(spinner='line')
stream_context = model.createStream()
wav_data = bytearray()
for frame in frames:
if frame is not None:
@ -185,25 +197,25 @@ def main(ARGS):
wav_data = bytearray()
text = model.finishStream(stream_context)
print("Recognized: %s" % text)
stream_context = model.setupStream()
stream_context = model.createStream()
if __name__ == '__main__':
BEAM_WIDTH = 500
DEFAULT_SAMPLE_RATE = 16000
LM_ALPHA = 0.75
LM_BETA = 1.85
N_FEATURES = 26
N_CONTEXT = 9
import argparse
parser = argparse.ArgumentParser(description="Stream from microphone to DeepSpeech using VAD")
parser.add_argument('-v', '--vad_aggressiveness', type=int, default=3,
help="Set aggressiveness of VAD: an integer between 0 and 3, 0 being the least aggressive about filtering out non-speech, 3 the most aggressive. Default: 3")
help="Set aggressiveness of VAD: an integer between 0 and 3, 0 being the least aggressive about filtering out non-speech, 3 the most aggressive. Default: 3")
parser.add_argument('--nospinner', action='store_true',
help="Disable spinner")
help="Disable spinner")
parser.add_argument('-w', '--savewav',
help="Save .wav files of utterences to given directory")
help="Save .wav files of utterences to given directory")
parser.add_argument('-f', '--file',
help="Read from .wav file instead of microphone")
parser.add_argument('-m', '--model', required=True,
help="Path to the model (protocol buffer binary file, or entire directory containing all standard-named files for model)")
@ -214,13 +226,9 @@ if __name__ == '__main__':
parser.add_argument('-t', '--trie', default='trie',
help="Path to the language model trie file created with native_client/generate_trie. Default: trie")
parser.add_argument('-d', '--device', type=int, default=None,
help="Device input index (Int) as listed by pyaudio.PyAudio.get_device_info_by_index(). If not provided, falls back to PyAudio.get_default_device()")
help="Device input index (Int) as listed by pyaudio.PyAudio.get_device_info_by_index(). If not provided, falls back to PyAudio.get_default_device().")
parser.add_argument('-r', '--rate', type=int, default=DEFAULT_SAMPLE_RATE,
help=f"Input device sample rate. Default: {DEFAULT_SAMPLE_RATE}. Your device may require 44100.")
parser.add_argument('-nf', '--n_features', type=int, default=N_FEATURES,
help=f"Number of MFCC features to use. Default: {N_FEATURES}")
parser.add_argument('-nc', '--n_context', type=int, default=N_CONTEXT,
help=f"Size of the context window used for producing timesteps in the input vector. Default: {N_CONTEXT}")
parser.add_argument('-la', '--lm_alpha', type=float, default=LM_ALPHA,
help=f"The alpha hyperparameter of the CTC decoder. Language Model weight. Default: {LM_ALPHA}")
parser.add_argument('-lb', '--lm_beta', type=float, default=LM_BETA,

View File

@ -1,5 +1,6 @@
deepspeech~=0.4.1
deepspeech~=0.6.0a5
pyaudio~=0.2.11
webrtcvad~=2.0.10
halo~=0.0.18
numpy~=1.15.1
scipy~=1.1.0

View File

@ -0,0 +1,21 @@
#!/bin/bash
set -xe
THIS=$(dirname "$0")
pushd ${THIS}
source ../tests.sh
pip install --user $(get_python_wheel_url "$1")
pip install --user -r requirements.txt
pulseaudio &
python mic_vad_streaming.py \
--model $HOME/DeepSpeech/models/output_graph.pbmm \
--alphabet $HOME/DeepSpeech/models/alphabet.txt \
--lm $HOME/DeepSpeech/models/lm.binary \
--trie $HOME/DeepSpeech/models/trie \
--file $HOME/DeepSpeech/audio/2830-3980-0043.wav
popd

View File

@ -6,19 +6,17 @@ const Duplex = require('stream').Duplex;
const Wav = require('node-wav');
const BEAM_WIDTH = 1024;
const N_FEATURES = 26;
const N_CONTEXT = 9;
let modelPath = './models/output_graph.pbmm';
let alphabetPath = './models/alphabet.txt';
let model = new DeepSpeech.Model(modelPath, N_FEATURES, N_CONTEXT, alphabetPath, BEAM_WIDTH);
let model = new DeepSpeech.Model(modelPath, alphabetPath, BEAM_WIDTH);
const LM_ALPHA = 0.75;
const LM_BETA = 1.85;
let lmPath = './models/lm.binary';
let triePath = './models/trie';
model.enableDecoderWithLM(alphabetPath, lmPath, triePath, LM_ALPHA, LM_BETA);
model.enableDecoderWithLM(lmPath, triePath, LM_ALPHA, LM_BETA);
let audioFile = process.argv[2] || './audio/2830-3980-0043.wav';
@ -69,4 +67,4 @@ audioStream.on('finish', () => {
let result = model.stt(audioBuffer.slice(0, audioBuffer.length / 2), 16000);
console.log('result:', result);
});
});

View File

@ -8,7 +8,7 @@
},
"dependencies": {
"argparse": "^1.0.10",
"deepspeech": "^0.4.1",
"deepspeech": "^0.6.0-alpha.5",
"node-wav": "0.0.2",
"sox-stream": "^2.0.3",
"util": "^0.11.1"

18
examples/nodejs_wav/test.sh Executable file
View File

@ -0,0 +1,18 @@
#!/bin/bash
set -xe
THIS=$(dirname "$0")
pushd ${THIS}
source ../tests.sh
npm install $(get_npm_package_url)
npm install
ln -s $HOME/DeepSpeech/models models
node index.js $HOME/DeepSpeech/audio/2830-3980-0043.wav
node index.js $HOME/DeepSpeech/audio/8455-210777-0068.wav
node index.js $HOME/DeepSpeech/audio/4507-16021-0012.wav
popd

23
examples/tests.sh Executable file
View File

@ -0,0 +1,23 @@
#!/bin/bash
set -xe
THIS=$(dirname "$0")
source ../../taskcluster/tc-tests-utils.sh
DEP_TASK_ID=$(curl -s https://queue.taskcluster.net/v1/task/${TASK_ID} | python -c 'import json; import sys; print(" ".join(json.loads(sys.stdin.read())["dependencies"]));')
get_python_wheel_url()
{
local this_python_version=$1
extract_python_versions "${this_python_version}" "pyver" "pyver_pkg" "py_unicode_type" "pyconf" "pyalias"
echo "$(get_python_pkg_url "${pyver_pkg}" "${py_unicode_type}" "deepspeech" https://queue.taskcluster.net/v1/task/${DEP_TASK_ID}/artifacts/public)"
}
get_npm_package_url()
{
echo "https://queue.taskcluster.net/v1/task/${DEP_TASK_ID}/artifacts/public/deepspeech-${DS_VERSION}.tgz"
}

View File

@ -22,10 +22,10 @@ def main(args):
parser.add_argument('--stream', required=False, action='store_true',
help='To use deepspeech streaming interface')
args = parser.parse_args()
if args.stream is True and len(sys.argv[1:]) == 3:
print("Opening mic for streaming")
elif args.audio is not None and len(sys.argv[1:]) == 6:
logging.debug("Transcribing audio file @ %s" % args.audio)
if args.stream is True:
print("Opening mic for streaming")
elif args.audio is not None:
logging.debug("Transcribing audio file @ %s" % args.audio)
else:
parser.print_help()
parser.exit()
@ -72,7 +72,7 @@ def main(args):
logging.debug("************************************************************************************************************")
print("%-30s %-20.3f %-20.3f %-20.3f %-0.3f" % (filename + ext, audio_length, inference_time, model_retval[1], model_retval[2]))
else:
sctx = model_retval[0].setupStream()
sctx = model_retval[0].createStream()
subproc = subprocess.Popen(shlex.split('rec -q -V0 -e signed -L -c 1 -b 16 -r 16k -t raw - gain -2'),
stdout=subprocess.PIPE,
bufsize=0)

View File

@ -283,7 +283,7 @@ class App(QMainWindow):
logging.debug("Start Recording pressed")
logging.debug("Preparing for transcription...")
sctx = self.model[0].setupStream()
sctx = self.model[0].createStream()
subproc = subprocess.Popen(shlex.split('rec -q -V0 -e signed -L -c 1 -b 16 -r 16k -t raw - gain -2'),
stdout=subprocess.PIPE,
bufsize=0)

View File

@ -1,3 +1,3 @@
deepspeech==0.4.1
deepspeech~=0.6.0a5
webrtcvad
pyqt5

View File

@ -0,0 +1,23 @@
#!/bin/bash
set -xe
THIS=$(dirname "$0")
pushd ${THIS}
source ../tests.sh
pip install --user $(get_python_wheel_url "$1")
pip install --user -r requirements.txt
python audioTranscript_cmd.py \
--audio $HOME/DeepSpeech/audio/2830-3980-0043.wav \
--aggressive 0 \
--model $HOME/DeepSpeech/models/
python audioTranscript_cmd.py \
--audio $HOME/DeepSpeech/audio/2830-3980-0043.wav \
--aggressive 0 \
--model $HOME/DeepSpeech/models/ \
--stream
popd

View File

@ -16,19 +16,17 @@ Load the pre-trained model into the memory
Returns a list [DeepSpeech Object, Model Load Time, LM Load Time]
'''
def load_model(models, alphabet, lm, trie):
N_FEATURES = 26
N_CONTEXT = 9
BEAM_WIDTH = 500
LM_ALPHA = 0.75
LM_BETA = 1.85
model_load_start = timer()
ds = Model(models, N_FEATURES, N_CONTEXT, alphabet, BEAM_WIDTH)
ds = Model(models, alphabet, BEAM_WIDTH)
model_load_end = timer() - model_load_start
logging.debug("Loaded model in %0.3fs." % (model_load_end))
lm_load_start = timer()
ds.enableDecoderWithLM(alphabet, lm, trie, LM_ALPHA, LM_BETA)
ds.enableDecoderWithLM(lm, trie, LM_ALPHA, LM_BETA)
lm_load_end = timer() - lm_load_start
logging.debug('Loaded language model in %0.3fs.' % (lm_load_end))

View File

@ -0,0 +1,61 @@
$if: '(event.event != "push") && (event.event != "tag")'
then:
taskId: ${taskcluster.taskId}
provisionerId: ${taskcluster.docker.provisionerId}
workerType: ${taskcluster.docker.workerType}
taskGroupId: ${taskcluster.taskGroupId}
schedulerId: ${taskcluster.schedulerId}
dependencies:
$map: { $eval: build.dependencies }
each(b):
$eval: as_slugid(b)
created: { $fromNow: '0 sec' }
deadline: { $fromNow: '1 day' }
expires: { $fromNow: '7 days' }
extra:
github:
{ $eval: taskcluster.github_events.pull_request }
routes:
- "notify.irc-channel.${notifications.irc}.on-exception"
- "notify.irc-channel.${notifications.irc}.on-failed"
scopes: [
"queue:route:notify.irc-channel.*"
]
payload:
maxRunTime: { $eval: to_int(build.maxRunTime) }
image: ${build.docker_image}
env:
DEEPSPEECH_MODEL: "https://github.com/lissyx/DeepSpeech/releases/download/tc-0.6.0/models.tar.gz"
DEEPSPEECH_AUDIO: "https://github.com/mozilla/DeepSpeech/releases/download/v0.4.1/audio-0.4.1.tar.gz"
PIP_DEFAULT_TIMEOUT: "60"
command:
- "/bin/bash"
- "--login"
- "-cxe"
- $let:
extraSystemSetup: { $eval: strip(str(build.system_setup)) }
in: >
apt-get -qq update && apt-get -qq -y upgrade && apt-get -qq -y install git sox sudo && ${extraSystemSetup} &&
adduser --system --home ${system.homedir.linux} ${system.username} &&
cd ${system.homedir.linux} &&
echo -e "#!/bin/bash\nset -xe\n env && id && mkdir ~/DeepSpeech/ && git clone --quiet ${event.head.repo.url} ~/DeepSpeech/ds/ && cd ~/DeepSpeech/ds && git checkout --quiet ${event.head.sha} && wget -O - $DEEPSPEECH_MODEL | tar -C ~/DeepSpeech/ -xzvf - && wget -O - $DEEPSPEECH_AUDIO | tar -C ~/DeepSpeech/ -xzvf - " > /tmp/clone.sh && chmod +x /tmp/clone.sh &&
sudo -H -u ${system.username} /bin/bash /tmp/clone.sh &&
sudo -H -u ${system.username} --preserve-env /bin/bash ${build.args.tests_cmdline}
artifacts:
"public":
type: "directory"
path: "/tmp/artifacts/"
expires: { $fromNow: '7 days' }
metadata:
name: ${build.metadata.name}
description: ${build.metadata.description}
owner: ${event.head.user.email}
source: ${event.head.repo.url}

View File

@ -0,0 +1,13 @@
build:
template_file: examples-base.tyml
docker_image: "node:10"
dependencies:
- "node-package-cpu"
system_setup:
>
apt-get -qq -y install ffmpeg
args:
tests_cmdline: "${system.homedir.linux}/DeepSpeech/ds/examples/ffmpeg_vad_streaming/test.sh"
metadata:
name: "DeepSpeech examples: ffmpeg VAD Streaming NodeJS v10.x"
description: "DeepSpeech examples: ffmpeg VAD Streaming NodeJS v10.x"

View File

@ -0,0 +1,13 @@
build:
template_file: examples-base.tyml
docker_image: "node:8"
dependencies:
- "node-package-cpu"
system_setup:
>
apt-get -qq -y install ffmpeg
args:
tests_cmdline: "${system.homedir.linux}/DeepSpeech/ds/examples/ffmpeg_vad_streaming/test.sh"
metadata:
name: "DeepSpeech examples: ffmpeg VAD Streaming NodeJS v8.x"
description: "DeepSpeech examples: ffmpeg VAD Streaming NodeJS v8.x"

View File

@ -0,0 +1,13 @@
build:
template_file: examples-base.tyml
docker_image: "python:3.6"
dependencies:
- "linux-amd64-cpu-opt"
system_setup:
>
apt-get -qq -y install portaudio19-dev pulseaudio
args:
tests_cmdline: "${system.homedir.linux}/DeepSpeech/ds/examples/mic_vad_streaming/test.sh 3.6.0:m"
metadata:
name: "DeepSpeech examples: mic VAD streaming Py3.6"
description: "DeepSpeech examples: mic VAD streaming Python 3.6"

View File

@ -0,0 +1,13 @@
build:
template_file: examples-base.tyml
docker_image: "python:3.7"
dependencies:
- "linux-amd64-cpu-opt"
system_setup:
>
apt-get -qq -y install portaudio19-dev pulseaudio
args:
tests_cmdline: "${system.homedir.linux}/DeepSpeech/ds/examples/mic_vad_streaming/test.sh 3.7.0:m"
metadata:
name: "DeepSpeech examples: mic VAD streaming Py3.7"
description: "DeepSpeech examples: mic VAD streaming Python 3.7"

View File

@ -0,0 +1,10 @@
build:
template_file: examples-base.tyml
docker_image: "node:10"
dependencies:
- "node-package-cpu"
args:
tests_cmdline: "${system.homedir.linux}/DeepSpeech/ds/examples/nodejs_wav/test.sh"
metadata:
name: "DeepSpeech examples: NodeJS WAV NodeJS v10.x"
description: "DeepSpeech examples: NodeJS WAV NodeJS v10.x"

View File

@ -0,0 +1,10 @@
build:
template_file: examples-base.tyml
docker_image: "node:8"
dependencies:
- "node-package-cpu"
args:
tests_cmdline: "${system.homedir.linux}/DeepSpeech/ds/examples/nodejs_wav/test.sh"
metadata:
name: "DeepSpeech examples: NodeJS WAV NodeJS v8.x"
description: "DeepSpeech examples: NodeJS WAV NodeJS v8.x"

View File

@ -0,0 +1,10 @@
build:
template_file: examples-base.tyml
docker_image: "python:3.5"
dependencies:
- "linux-amd64-cpu-opt"
args:
tests_cmdline: "${system.homedir.linux}/DeepSpeech/ds/examples/vad_transcriber/test.sh 3.5.0:m"
metadata:
name: "DeepSpeech examples: VAD transcriber Py3.5"
description: "DeepSpeech examples: VAD transcriberaming Python 3.5"

View File

@ -0,0 +1,10 @@
build:
template_file: examples-base.tyml
docker_image: "python:3.6"
dependencies:
- "linux-amd64-cpu-opt"
args:
tests_cmdline: "${system.homedir.linux}/DeepSpeech/ds/examples/vad_transcriber/test.sh 3.6.0:m"
metadata:
name: "DeepSpeech examples: VAD transcriber Py3.6"
description: "DeepSpeech examples: VAD transcriberaming Python 3.6"

View File

@ -0,0 +1,10 @@
build:
template_file: examples-base.tyml
docker_image: "python:3.7"
dependencies:
- "linux-amd64-cpu-opt"
args:
tests_cmdline: "${system.homedir.linux}/DeepSpeech/ds/examples/vad_transcriber/test.sh 3.7.0:m"
metadata:
name: "DeepSpeech examples: VAD transcriber Py3.7"
description: "DeepSpeech examples: VAD transcriberaming Python 3.7"