voice_segmentor

A set of scripts that reads media -> detect voice from the media -> create wav for each sentences

Scripts

folder_convert_media_to_wav.py

convert all non-wav media into wav in a given folder it uses ffmpeg internally as that's performing the best (most stable, and compatible with any) ffmpeg uses default which is 16 bit PCM, 48Khz, 2ch

Usage:

python folder_convert_media_to_wav.py -f <folder name>

---

segment-media-vad-onx.py

Reads either a single wave file or all the wave files in a specific folder, and run Voice Activity Dector (with onx option enabled) to generate a folder that has each sentences as wave file it internally uses https://github.com/snakers4/silero-vad for Voice activity Detection

Usage: For a single wave file

python segment-media-vad-onx.py <wave file>

For all wave file in a folder

python segment-media-vad-onx.py -f <folder>

---

labeling-with-whipser.py

Read a folder that has list of wave files (assuming it's a segmented wav files by segment-media-vad-onx.py`,

1. create a excel file that has the same name as the folder
2. add header row of Filename (first column), Transcription (second column) to the first row
3. for the first column, fill up all the rows withe the wav file names
4. for the second column, fill up all the rows by running whisper to the corresponding wav file
5. once done, save that to the excel file

Usage: For all wave file in a folder

python labeling-with-whipser.py -f <folder>