Scripts for augmenting speech dataset using multiple techniques
Install ffmpeg before trying the scripts.
There are total 7 scripts to augment speech data (in order of priority).
-
mp32wav.sh- It can be used to convert .mp3 files to .wav files as .wav files are used widely.
Usage : ./mp32wav.sh InputFolder(mp3) OutputFolder(wav) -
speedAdjustWav.sh- It can be used to augment the speed of files.
Usage : ./speedAdjustWav.sh InputFolder OutputFolder Speed
(Note: For best results, use "Speed" values from [0.7,1.4] only) -
volumeAdjustWav.sh- It can be used to adjust the amplitude or volume of the speech. This is useful in enhancing robustness to distance of speaker.
Usage : ./volumeAdjustWav.sh InputFolder OutputFolder -
equalTime.sh- It can be used to adjust to make the size of each .wav file equal to (atleast) MinimumSeconds.
Usage : ./equalTime.sh InputFolder OutputFolder MinimumSeconds -
mergeNoise.sh- It can be used to mix noise in the dataset by superimposing the two streams.
Usage : ./mergeNoise.sh InputFolder NoiseFolder OutputFolder -
splitAudioSegment.sh- It can be used to split large .wav files into small fragments of equal size(s).
Usage : ./splitAudioSegment.sh InputFolder SegmentDuration(in seconds) -
renameSeq.sh- It can be used to rename all the files in a folder. All files will be renamed in a particular sequence.
Usage : ./renameSeq.sh InputFolder Phrase
(Note: Phrase can be any alphanumeric string which can be used to differentiate between data from different users)