soft fog
This is a tweaked version of jarredou's colab notebook that clones and runs ZFTurbo's MSST repository.
Key Features
- Separation
- Demudder - Phase Remix (Similar to X-Minus): This method is only recommended for models that produce very muddy instrumentals!
- Phase Fixer: Copies FFT phase of the source audio to the target audio within frequency cutoff range. Particularly useful for removing noise/vocal bleed on unwa's v1 or becruily's instrumental model stem.
- Ensemble: Processes an output audio as min/max/avg of waveform or FFT magnitude of several input audios.
- Zip/Unzip
Note
- If there are models you want but not included in the colab, you can leave me suggestion on it or edit the code in Separation cell. Any other feedbacks and issue report are welcome.
- Newly added models and some updates are at comments below.
Try it now!
