pipeline.py

import os
import glob
import shutil
from audiosplitter import split_all_audios, split_long_audios, filter_short_audios
import argparse

# clear out the wavs folder using python

# if wavs folder doesn't exist, create it
if not os.path.exists('./wavs'):
    os.makedirs('./wavs')

def clear_wavs():
    for wav in glob.glob('./wavs/*.wav'):
        os.remove(wav)

# copy all wavs from put_audio_files_here into wavs
def copy_wavs():
    for wav in glob.glob('./put_audio_files_here/*.wav'):
        shutil.copy(wav, './wavs/')




def main():
    parser = argparse.ArgumentParser(
        prog='LJ Speech Toolset',
        description='chops up wav files and adds transcription, outputs in LJ Speech format')
    parser = argparse.ArgumentParser()
    parser.add_argument('-p', '--provider',  help='Set transcription provider (google or whisper) default is whisper', default="whisper")
    parser.add_argument('-a', '--speech_key',  help='Google Speech API Key')
    parser.add_argument('-m', '--model',   help='Open AI Whisper model (tiny, base, small, medium, large, large-v2, or large-v3) to use, default large-v3', default="large-v3")
    parser.add_argument('-s', '--seconds',  help='Set the number seconds per wav file.', default=12)
    parser.add_argument('-l', '--min_silence_len',  help='Set the min_silence_len for audio segmenting', default=300)
    parser.add_argument('-t', '--silence_thresh',  help='Set silence_thresh for audio segmenting', default=-60)
    parser.add_argument('-k', '--keep_silence',  help='Set keep_silence for audio segmenting', default=300)
    parser.add_argument('-n', '--no_split',  action="store_true", help='Do not split the metadata.csv into training and test sets', default=300)

    args = parser.parse_args()

    if args.provider == "google" and args.speech_key == None:
        print("Error: you must provide a google speech api key to use google transcriber")
        return



    clear_wavs();
    #not required, leaving for historical purposes.  Original author can remove
#    copy_wavs();

    # 1. split audio files with audiosplitter

    # not required, leaving for historical purposes.  Original author can remove
#    split_all_audios();

    # simply split and move the clips
    split_long_audios(min_silence_len = args.min_silence_len ,silence_thresh = args.silence_thresh, keep_silence=args.keep_silence, seconds = args.seconds);
    filter_short_audios();

    # 2. transcribe audio files with transcriber

    from transcriber import transcribe

    transcribe(provider=args.provider, google_speech_api_key=args.speech_key, model_name=args.model);

    # 3. swearing

    from swearing import replace_censored_words

    replace_censored_words();

    print('replace_censored_words')

    # 4. prepare dataset

    from make_dataset import make_dataset

    make_dataset(args.no_split);

    print('ok')

    # delete wavs_split_temp
    shutil.rmtree('wavs_split_temp')

    # delete wavs_split_final
    shutil.rmtree('wavs_split_final')

    # delete wavs
    clear_wavs();


if __name__ == "__main__":
    main()