3 ms·
Is there a paper on how they're resynthesizing phase components? IME neutral networks are real real bad at handling fft phase, so separation tend to use frequen
by sdenton4 5y ago
Is there a paper on how they're resynthesizing phase components? IME neutral networks are real real bad at handling fft phase, so separation tend to use frequency making, or use a learned filter bank.
- gavinray 5y agoThere's maybe something useful for you here? https://github.com/facebookresearch/demucs https://github.com/facebookresearch/demucs https://github.com/sigsep/open-unmix-pytorch https://github.com/sigsep/open-unmix-pytorch https://github.com/bytedance/music_source_separation https://github.com/bytedance/music_source_separation https://github.com/deezer/spleeter https://github.com/deezer/spleeter
- sdenton4 5y agoLooks like demucs is using a learned filter bank (so no handling of fft phase), and open umix and spleeter are using spectrogram magnitude masking (just reuse phase from original fft in all channels). Can't immediately tell what bytedance is doing.