E-DAIC dataset preprocess

This is a simple project to clean and filter the original files in E-DAIC dataset, the original data struture should be:

-301_P
    -301_Transcript.csv
    -310_AUDIO.wav

steps:

• step 1: change the data structure easier to view and process like this

-trans
    -301.json
    -302.json
    -...

-audio
    -301.wav
    -302.wav
    -...

• step 2: filter out the interviewer's voice using DeepFilterNet.

OR simply use Whisper to transcrible the audio files -,-

• step 3.1: segment audio files using original json files, automatically correct errors like starting time is bigger than ending time, ending time is bigger than the audio length.

• step 3.2: feed each segment to whisper and replace the original text with new ones.

usage