mirror of
https://github.com/TaterTotterson/microWakeWord-Trainer-Nvidia-Docker.git
synced 2026-06-12 20:10:19 -06:00
cb81f7f02d
The files in the `cli` directory allow you to train wake words from the command line without needing to use the Jupyter notebook or a web browser. Basically, the logic from the notebook has been placed in separate shell scripts and python files wrapped by 3 high-level scripts that do the following: * setup_python_venv: Creates a Python virtual environment with all the packages needed to train. The venv is created in the container's /data directory and is therefore stored on the host, not in the container's root docker volume. * setup_training_datasets: Downloads, extracts and converts the MIT RIR, FMA, Audioset and Negative training reference datasets. Also stored in /data. * train_wake_word: Generates the wake word samples, augments them with the audio from the training datasets, and finally runs the microwakeword training. The resulting model tflite and json files are placed in the /data/output directory. See the README.md file for much more information.
113 lines
3.5 KiB
Bash
Executable File
113 lines
3.5 KiB
Bash
Executable File
#!/bin/bash
|
|
set -e
|
|
|
|
PROGPATH=$(realpath "$0")
|
|
PROGDIR=$(dirname "${PROGPATH}")
|
|
|
|
KNOWN_ARGS=( samples batch-size data-dir )
|
|
source "${PROGDIR}/shell.functions"
|
|
WAKE_WORD="${POSITIONAL_ARGS[0]}"
|
|
|
|
if [ ${#UNKNOWN_ARGS[@]} -gt 0 ] ; then
|
|
echo "Unknown argument(s): ${UNKNOWN_ARGS[*]}" >&2
|
|
HELP=true
|
|
fi
|
|
|
|
if [ "${HELP}" == "true" ] || [ -z "${WAKE_WORD}" ] ; then
|
|
cat <<EOF >&2
|
|
Usage: $0 [ --samples=<samples> ] [ --batch-size=<batch_size> ] <wake_word>
|
|
|
|
--samples: The number of samples to generate for the wake word.
|
|
Default: ${DEFAULT_SAMPLES}
|
|
|
|
--batch-size: How many samples should be generated at a time. The more
|
|
samples, the more memory is needed.
|
|
Default: ${DEFAULT_BATCH_SIZE}
|
|
|
|
<wake_word> The word to generate samples for.
|
|
Required.
|
|
|
|
EOF
|
|
exit 1
|
|
fi
|
|
|
|
# shellcheck source=/dev/null
|
|
source "${DATA_DIR}/.venv/bin/activate"
|
|
|
|
WORK_DIR="${DATA_DIR}/work"
|
|
mkdir -p "${WORK_DIR}" || :
|
|
cd "${WORK_DIR}"
|
|
|
|
PSG="${DATA_DIR}/tools/piper-sample-generator"
|
|
MODELS_DIR="${PSG}/models"
|
|
MODEL_NAME=en_US-libritts_r-medium.pt
|
|
MODEL_FILE="${MODELS_DIR}/${MODEL_NAME}"
|
|
SAMPLES_DIR="${WORK_DIR}/wake_word_samples"
|
|
|
|
mkdir -p "${SAMPLES_DIR}" || :
|
|
|
|
REGENERATE=false
|
|
|
|
if [ "${SAMPLES}" -eq 1 ] ; then
|
|
echo "===== Generating ${SAMPLES} sample of '${WAKE_WORD}' ====="
|
|
wake_word_filename="${WAKE_WORD//[ \`~\!\$&*\(\)\{\}\[\]\|\;\'\"<>.?\/]/_}"
|
|
|
|
mkdir -p "${WORK_DIR}/test_sample" || :
|
|
"${PSG}/generate_samples.py" "${WAKE_WORD}" \
|
|
--model "${MODEL_FILE}" \
|
|
--max-samples ${SAMPLES} \
|
|
--batch-size ${BATCH_SIZE} \
|
|
--output-dir "${WORK_DIR}/test_sample" \
|
|
--max-speakers 100 2>&1 | sed -r -e "s/(DEBUG|INFO):__main__:/ /g"
|
|
mv "${WORK_DIR}/test_sample/0.wav" "${WORK_DIR}/test_sample/${wake_word_filename}.wav"
|
|
echo "Sample available at ${WORK_DIR}/test_sample/${wake_word_filename}.wav"
|
|
echo "Play it from your host."
|
|
exit 0
|
|
fi
|
|
|
|
grep -q "${WAKE_WORD}:${SAMPLES}:${MODEL_NAME}" "${WORK_DIR}/last_wake_word" &>/dev/null || REGENERATE=true
|
|
|
|
# Double check that the number of existing samples matches SAMPLES"
|
|
existing_samples=$(find "${SAMPLES_DIR}" -name '*.wav' | wc -l)
|
|
[ "${existing_samples}" -eq "${SAMPLES}" ] || REGENERATE=true
|
|
|
|
START_TS=$EPOCHSECONDS
|
|
|
|
if ! ${REGENERATE} ; then
|
|
echo "Sample generation not required"
|
|
echo
|
|
exit 0
|
|
fi
|
|
|
|
echo -e "\n===== Generating ${SAMPLES} wake word samples in batches of ${BATCH_SIZE} ====="
|
|
export TF_CPP_MIN_LOG_LEVEL=9
|
|
export TF_FORCE_GPU_ALLOW_GROWTH=true
|
|
export TF_GPU_ALLOCATOR=cuda_malloc_async
|
|
export TF_XLA_FLAGS="--tf_xla_auto_jit=0"
|
|
export NVIDIA_TF32_OVERRIDE=1
|
|
export TF_CUDNN_WORKSPACE_LIMIT_IN_MB=512
|
|
export GLOG_minloglevel=2
|
|
export GRPC_VERBOSITY=ERROR
|
|
|
|
echo " Generating samples"
|
|
rm -rf "${SAMPLES_DIR}" || :
|
|
mkdir -p "${SAMPLES_DIR}" || :
|
|
"${PSG}/generate_samples.py" "${WAKE_WORD}" \
|
|
--model "${MODEL_FILE}" \
|
|
--max-samples ${SAMPLES} \
|
|
--batch-size ${BATCH_SIZE} \
|
|
--output-dir "${SAMPLES_DIR}" 2>&1 | sed -r -e "s/(DEBUG|INFO):__main__:/ /g"
|
|
|
|
generated_files=$(find "${SAMPLES_DIR}" -name '*.wav' | wc -l)
|
|
if [ "${generated_files}" -ne "${SAMPLES}" ] ; then
|
|
echo "ERROR: only generated ${generated_files} files" >&2
|
|
exit 1
|
|
fi
|
|
END_TS=$(date +%s.%N)
|
|
echo "${WAKE_WORD}:${SAMPLES}:${MODEL_NAME}" > "${WORK_DIR}/last_wake_word"
|
|
echo
|
|
END_TS=$EPOCHSECONDS
|
|
print_elapsed_time "${START_TS}" "${END_TS}" "Generated ${SAMPLES} wake word samples."
|
|
|
|
exit 0
|