Which model is better?
Added 2024-11-05 08:27:01 +0000 UTCThe Mel-RoFormer model by unwa v2 has just been added! We're asking you to test it out and vote.
Comments
is there a way to lay my vocals on a track?
willy jack
2025-04-20 17:25:18 +0000 UTCБЛин я ж оплатил
Жалгас Есенбаев
2025-02-27 11:01:58 +0000 UTCПочему не работает
Жалгас Есенбаев
2025-02-27 11:01:35 +0000 UTCDo you have something coming were it can separate 2 vocals?
Knowledge Bey
2025-01-15 00:59:33 +0000 UTCWhen another model will come ? Im really impatient!! Unwa v2 is not bad but I really want better
Semih Acer
2024-12-15 19:09:08 +0000 UTCMy method is to use one of these, depending on the track's quality. For example, v1 gives much better results for MP3s, while v2 is better for lossless originals. Then I put the result into the denoiser.
Tim Shikolenkov
2024-11-26 19:24:49 +0000 UTCDon't expect wonders. Melroformers are the BEST models out there, no doubt about it. FB338 was (for me) the best before BS and Mel appeared. Now I only use FB338, when and if I use it, for that scenario I said before, small parts with quiet instruments
Cristobal Ramos
2024-11-06 09:34:07 +0000 UTCv1 and v2 with postprocessing
Raffaela
2024-11-06 02:15:20 +0000 UTCHave you tested v1 or v2?
deleted_user_123
2024-11-06 01:09:24 +0000 UTCThanks for sharing, looking forward to trying it out!
Gatto
2024-11-06 00:44:42 +0000 UTCHere it is, Gatto. Just drop it in the models/MDX_Net_Models folder in the Ultimate Vocal Remover installation folder, it should show up in the list of models when you start the program https://drive.google.com/file/d/1zIqoCHQyYQHz7F9mib43_vuM5psobiTO/view?usp=sharing
Cristobal Ramos
2024-11-05 19:40:08 +0000 UTCAs I already said on the channel, in my opinion Kim's model is far better: the Unwa model leaves a very annoying background noise, despite the sound quality being similar.
Raffaela
2024-11-05 19:09:33 +0000 UTCSure man, no rush, take your time :)
Gatto
2024-11-05 16:47:16 +0000 UTCGatto, sorry, I'm still working on it (actually, I'm writing this in my laptop because the OS migration in my desktop PC went like cr*p), but the moment I can I'll send you that link
Cristobal Ramos
2024-11-05 16:36:02 +0000 UTCWould be really appreciated! I also find myself with some faint vocal leftovers in a few tracks the MelRoFormer models couldn't fully isolate, so it's worth giving a shot to other models for those specific parts.
Gatto
2024-11-05 11:37:10 +0000 UTCI downloaded it and added it to UVR (Ultimate Vocal Remover 5.6.0). You need that program to run it. Overall, the Melroformers are much better, but in some songs they still leave faint traces of the voice in the instrumental, in quiet parts. FB338 takes better care of those parts without mudding the sound, though it may add some noise, so I just use it specifically for those parts, while for the rest of the song I go with melroformer. Give me an hour (I'm migrating my OS to a new drive) and I'll link you the FB338 model to download
Cristobal Ramos
2024-11-05 11:21:19 +0000 UTCWhere do you access that FB338 model? I'm interested in trying it out.
Gatto
2024-11-05 10:55:30 +0000 UTCWhy choose? I would, but if that means removing one or the other I won't. The KIM model has the Demudder post-process which, depending on the song, gives better results than UNWA2, plus Max+Mag Drums. The combo Demudder/MMDrums is a killer UNWA2 doesn't seem to be that major an improvement (at least in the songs I've tried). I'm still using FB338 in some songs because, in some parts, still works better than the new models, and was removed from site. I wouldn't like the same happening to any of the Mel-Roformers at all
Cristobal Ramos
2024-11-05 10:14:53 +0000 UTCBoth seem to sound equivalent if I'm honest. What made v1 special is the amount of muddiness it eliminated from the old model, hopefully it'll be a step forward in that regard in the next versions in the future.
Gatto
2024-11-05 10:05:36 +0000 UTC
