Tool transforms audio-korpora of given formats into desired output-formats. Audio-Corpus could be M-AILABS, Apache CommonVoice or others. Please consult the changelog to see which corpora are currently supported.

As the software-package is not yet available on pypi build the wheel yourself:

• Have Python 3.7.3 or higher installed
• Have pip installed
• Have wheel-support installed (if not use: pip install wheel)

• Build the wheel:

  python setup.py bdist_wheel

Then install like any wheel:

pip install dist/audio_korpora_pipeline-0.10-py2.py3-none-any.whl

This tool does not automatically download corpora from the internet, as download links will change over time anyway. This means you have to get the data yourself and adjust the configuration according your filepaths.

The following is an example how to convert CommonVoice (Input) to M-AILABS and LJSpeech (Output) Example command:

audio_korpora_pipeline -c config.cfg --input_corpora="CommonVoice" --output_corpora="LJSpeech"

Other Example: Create one Fairseq-formatted output from three different datasets:

audio_korpora_pipeline -c config.cfg --input_corpora="Archimob,ChJugendsprache,UntranscribedVideo" --output_corpora="FairseqWav2Vec"

audio_korpora_pipeline.py

change configuration according your needs within config.cfg

nohup command &:

nohup audio_korpora_pipeline -c ~/datasets/audio-korpora-pipeline-config.cfg --input_corpora='Archimob,UntranscribedVideo,ChJugendsprache' --output_corpora='FairseqWav2Vec' &

Tail logs:

tail -n 50 -f ~/repositories/audio-korpora-pipeline/log.log

Still running?:

htop