search

Audio-AGI / AudioSep

Official implementation of "Separate Anything You Describe"
https://audio-agi.github.io/Separate-Anything-You-Describe/
MIT License
1.63k stars 118 forks source link

Requesting Code reference for metrics calculation #5

Closed stonelazy closed 1 year ago

stonelazy commented 1 year ago

Team, I was wondering if you haven't included code reference for evaluation metrics such as CSIG,CBAK, COVL, SSNR etc.. ? Do you have any plans for this please ?

liuxubo717 commented 1 year ago

The code is already here: https://github.com/Audio-AGI/AudioSep/blob/main/utils.py

I'll release the evaluation benchmark soon!

stonelazy commented 1 year ago

Thanks for the revert. Am sorry, but i couldn't see evaluation code for the these metrics specifically CSIG,CBAK,COVL,SSNR.

liuxubo717 commented 1 year ago

Thanks for the reminder, Will update it ASAP.

liuxubo717 commented 1 year ago

For speech enhancement metrics, I use pesq and pysepm packages. For example:

pesq import pesq import pysepm

pesq_ = pesq(sampling_rate, clean_wav, separated_wav, 'wb') (csig, cbak, covl) = pysepm.composite(clean_wav, separated_wav, sampling_rate) ssnr = pysepm.SNRseg(clean_wav, separated_wav, sampling_rate)