speech_reverb

This repository includes scripts for reverberation WAV file creation.

You can play back by clicking the following WAV file names.

fv01_t01_s01.wav	rev1_fv01_t01_s01.wav
rev2_fv01_t01_s01.wav	rev3_fv01_t01_s01.wav

Dependency

kaldi, Python2.7, sox

Only wav-reverberate is used here but several kaldi utils are running in back-end. Future work would be bailing out necessary parts.

Command

./run.sh <input file including WAV file paths per line>

For example,

head wavfilst.lst

/home/kwon/copora/NIKL_trim_head100ms/fv01/fv01_t01_s01.wav
/home/kwon/copora/NIKL_trim_head100ms/fv01/fv01_t01_s02.wav
/home/kwon/copora/NIKL_trim_head100ms/fv01/fv01_t01_s03.wav
/home/kwon/copora/NIKL_trim_head100ms/fv01/fv01_t01_s04.wav
/home/kwon/copora/NIKL_trim_head100ms/fv01/fv01_t01_s05.wav
/home/kwon/copora/NIKL_trim_head100ms/fv01/fv01_t01_s06.wav
/home/kwon/copora/NIKL_trim_head100ms/fv01/fv01_t01_s07.wav
/home/kwon/copora/NIKL_trim_head100ms/fv01/fv01_t01_s08.wav
/home/kwon/copora/NIKL_trim_head100ms/fv01/fv01_t01_s09.wav
/home/kwon/copora/NIKL_trim_head100ms/fv01/fv01_t01_s10.wav

./run.sh wavfile.lst

ls data/output_wav/ -1

rev1_fv01_t01_s01.wav
rev1_fv01_t01_s02.wav
rev1_fv01_t01_s03.wav
rev1_fv01_t01_s04.wav
rev1_fv01_t01_s05.wav
rev1_fv01_t01_s06.wav
rev1_fv01_t01_s07.wav
rev1_fv01_t01_s08.wav
rev1_fv01_t01_s09.wav
rev1_fv01_t01_s10.wav
rev2_fv01_t01_s01.wav
rev2_fv01_t01_s02.wav
rev2_fv01_t01_s03.wav
rev2_fv01_t01_s04.wav
rev2_fv01_t01_s05.wav
rev2_fv01_t01_s06.wav
rev2_fv01_t01_s07.wav
rev2_fv01_t01_s08.wav
rev2_fv01_t01_s09.wav
rev2_fv01_t01_s10.wav
rev3_fv01_t01_s01.wav
rev3_fv01_t01_s02.wav
rev3_fv01_t01_s03.wav
rev3_fv01_t01_s04.wav
rev3_fv01_t01_s05.wav
rev3_fv01_t01_s06.wav
rev3_fv01_t01_s07.wav
rev3_fv01_t01_s08.wav
rev3_fv01_t01_s09.wav
rev3_fv01_t01_s10.wav

Configuration

run.sh includes several variables to control reverberation WAV file creation. Here is a few information

• num_data_reps : the number of reverberation wav files for creation

• foreground/background_snrs : signal-to-noise ratios when additive noise is applied

• db_string : RIR (Room Impulse Response) databases used for the reverberation wav file creation You may want to try aalto, c4dm, mardy and varechoic which are not tested yet.

• base_rirs="simulated" : simulated RIRs are applied, otherwise real isotropic RIR is applied

Miscellaneous

kaldi has made use of reverberate_data_dir.py to be working for pre-trained phone moldes where phone alignments are obtained from clean speech. run.sh follows wav.scp structure and utils/reverberate_data_dir.py is modified to create reverberation WAV files.