search

Anjok07 / ultimatevocalremovergui

GUI for a Vocal Remover that uses Deep Neural Networks.
MIT License
17.26k stars 1.29k forks source link

Which model is better for extracting human voice without harmonies? #555

Open imagine725 opened 1 year ago

imagine725 commented 1 year ago

i want a pure vocal！！！

daddyisherekitten commented 1 year ago

i used vocal 1 not sure if its the best tho?

RavenV-tiff commented 1 year ago

My work flow is to separate the instrumental and vocal using ensemble avg/avg then take the vocal track of that and run the MDX karaoke 1/2 models. The vocal separation should leave the main vocals in theory. It can have artifacts since there really isn't a lot of models that separate backing vocals.

Elia1995-Official commented 1 year ago

Which model(s) do you use with ensemble avg/avg?

RavenV-tiff commented 1 year ago

I use UVR-MDX inst HQ 2 + inst main + inst_full_292 + 9482 + htdemucs. All the MDX use htdemucs as a secondary model. I will say recently I've moved to using MVSEP (https://github.com/ZFTurbo/MVSEP-MDX23-music-separation-model) as it produces the best quality separation I've found so far. Just speculation but that model and others from the MDX23 challenge will eventually be added to UVR

jarredou commented 1 year ago

Just speculation but that model and others from the MDX23 challenge will eventually be added to UVR

SAMI-ByteDance team don't know if they can/will release their 1st place code&models. 2nd place will not share.. 3, 4 and others have already shared their stuff (MVSep, Kimberly Jensen, KUIElab...)

RavenV-tiff commented 1 year ago

Yeah, I read about it shortly after posting. Official agreement of the challenge, for any prizes, were to have the model open-source after it was over, so hopefully those that can't/won't were disqualified from that. Either way, new and improved methods are always on the horizon.

T-oasterO-ven commented 1 year ago

Yeah, I read about it shortly after posting. Official agreement of the challenge, for any prizes, were to have the model open-source after it was over, so hopefully those that can't/won't were disqualified from that. Either way, new and improved methods are always on the horizon.

Sadly, that will very likely happen, that is if the CCP is involved/influencing as they are with many things coming through that Orwellian great firewall.