search

deezer / spleeter

Deezer source separation library including pretrained models.
https://research.deezer.com/projects/spleeter.html
MIT License
25.95k stars 2.84k forks source link

[Feature] Using provided instrumental for enhanced vocals separation #634

Closed LilyStilson closed 3 years ago

LilyStilson commented 3 years ago

Description

What about allowing user to optionally provide an instrumental track for Spleeter to work on? I don't know if it's possible or not, but I think that it would better understand what is going vocals, and it doesn't need to separate accompaniment, because it's already provided.

Additional information

Currently Adobe Audition can do sort of the same thing (using Sound Model and Sound Remover effect), but it relies on calculations and some in-house algorithm. I think. if there is will be AI involved in process, then vocals separation could be much more accurate and Spleeter won't confuse vocals for instruments.

romi1502 commented 3 years ago

Hi @LilyStilson, Indeed adding extra information from the available instrumental track could help separating the vocal track. However, this would require quite a different architecture and retraining models to support this feature, so we probably won't include anything like this in spleeter in the near future. Thanks for the suggestion!