search

facebookresearch / demucs

Code for the paper Hybrid Spectrogram and Waveform Source Separation
MIT License
8.35k stars 1.06k forks source link

Audio Configuration for Optimal Results #592

Closed csetanmayjain closed 7 months ago

csetanmayjain commented 7 months ago

❓ Audio Configuration for Optimal Results

I used to have audio with different sample rates and channels. I would like to know the best configuration that I should set to get the best results, such as the number of channels, sample rate, and bit rate. Do you have any suggestions on this, and does considering it make a difference in the results or not? Thanks.

CarlGao4 commented 7 months ago

Unless you train your own model, official models will only output dual channels, 44100Hz. It is suggested to input dual-channel audio only.

csetanmayjain commented 7 months ago

Thanks. Also, would it be advisable to upsample/downsample the audio to 44100 Hz to get the best results, or can we expect the same results with a different sample rate?

CarlGao4 commented 7 months ago

It will resample the audio to match the model before separating

csetanmayjain commented 7 months ago

Thanks :)