search

xiph / rnnoise

Recurrent neural network for audio noise reduction
BSD 3-Clause "New" or "Revised" License
3.97k stars 890 forks source link

How to evaluate the performance of denoising when i Compared two weights? #148

Open wangohaha opened 3 years ago

wangohaha commented 3 years ago

i trained two weights of the rnnoise. when i tested the wav by myself, each ones all got a good denoising performance. However , i do not know which one is the best ? so, do you know some ways to evaluate the performance of denoising or the code to evaluate the performance of denoising wavs?

jagger2048 commented 3 years ago

Hi, You can use the PESQ to evaluate the output of the two weights. python-pesq

B&R

wangohaha commented 3 years ago

ok,3q however when i use pypesq , i had a problem of the usage; the ref signal is the data before mixing noise or the data after mixing noise

guishengzhang commented 3 years ago

The ref signal should be clean speech before mixing noise.

RXAldreezee commented 3 years ago

Hi, You can use the PESQ to evaluate the output of the two weights. python-pesq

B&R

I agree with this. Make sure you have clean wav file and rnnoise-processed wav file. Produce a wav file in each of the model and use the python-pesq to compare. All wav files should have the same audio duration and 16k sampling frequency.

If its doesn't work, try https://github.com/schmiph2/pysepm