Utilities for preprocessing the Switchboard and WSJ corpora in Python3

• 2020.07.31 : wTIMIT support

Before using the utilities, some requirements must meet first:

• Install Python packages:

  tqdm
  torchaudio
  

• Install sph2pipe executable program

1. Convert .sph files in LDC2002S09-Hub5e_00 to .wav files.

   python3 sph2wav.py .sph <path to sph2pipe> <path to LDC2002S09-Hub5e_00/english> SWB
   

2. Split the eval2000 set by the rules in hub5e_00.pem.

   python3 swb_eval_splitter.py <path to LDC2002S09-Hub5e_00/english>
   

Convert .wv1 files in WSJ0 and WSJ1 to .wav files.

python3 sph2wav.py .wv1 <path to sph2pipe> <path to WSJ0/WSJ1> WSJ

Convert .WAV files in wTIMIT to .wav files.

python3 sph2wav.py .WAV <path to sph2pipe> <path to wTIMIT> wTIMIT