wsj0-convert

A Python script to convert the WSJ0 speech corpus to more friendly file formats.

Requirements

• sph2pipe in PATH. Get it from here.
• ffmpeg in PATH (not required if using the --no-flac option)

Usage

Simply run:

python wsj0_convert.py <path-to-WSJ0> <output-dir>

This will create an audio directory inside <output-dir> containing all the audio files in .flac format, or in .wav format if using the --no-flac option. The audio files are organized in sub-directories by speakers.

Notes

• Only .wv1 files are converted. .wv2 files are skipped.
• The original folder structure is not respected. Output files are organized in sub-directories by speaker.
• Total output size if 3.9 GB when using .flac format.
• For extra speaker information (e.g. gender), see here.