zhaoyi2/audio_augment

A tool/script for batch speech data enhancement with speed/volume/RIRS/MUSAN

Audio augmentation

(README.md)

Audio augment is a tool/script for batch audio data augmenation through speed, volume, reverb, noise based on kaldi and sox.

Installation

• kaldi
• sox
• cd audio_augment
• cd tools; make KALDI=/your_kaldi_path

Usage

• cd audio_augment
• vim run_aug.sh to change your input_path and out_path and save
• bash run_aug.sh, so easy!

Workflow(run_aug.sh)

• Stage 1: Data Preparation contain data and text
• Stage 2: speed 0.9/1.1
• Stage 3: volume +-db
• Stage 4: reverberation(RIRS)
• Stage 5: MUSAN(noise/music/babble)
• Stage 6: combine above data and select a subset of the augmend data list about twice the origin data
• Stage 7: data and label generation

generated Examples

• cd data/wav/train_aug listen a few enhanced aishell1 audio example through speed/volume/RIRS/MUSAN.

Reference

• MUSAN http://www.openslr.org/17/
• RIRS http://www.openslr.org/28/
• https://github.com/linan2/add_reverb2