STT/bin/run-ci-ldc93s1_new_bytes_tflite.sh

27 lines
788 B
Bash
Executable File

#!/bin/sh
set -xe
ldc93s1_dir="./data/smoke_test"
ldc93s1_csv="${ldc93s1_dir}/ldc93s1.csv"
audio_sample_rate=$1
if [ ! -f "${ldc93s1_dir}/ldc93s1.csv" ]; then
echo "Downloading and preprocessing LDC93S1 example data, saving in ${ldc93s1_dir}."
python -u bin/import_ldc93s1.py ${ldc93s1_dir}
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 train.py --show_progressbar false \
--n_hidden 100 \
--checkpoint_dir '/tmp/ckpt_bytes' \
--export_dir '/tmp/train_bytes_tflite' \
--scorer_path 'data/smoke_test/pruned_lm.bytes.scorer' \
--bytes_output_mode true \
--audio_sample_rate ${audio_sample_rate} \
--export_tflite true