Vocal & Instrumental Isolation

The MVSep Drums model exists in 3 different variants based on following architectures: HTDemucs4, MelRoformer and SCNet. The model produces high-quality separation of music into a drums part and everything else.

Quality metrics

Algorithm name	Multisong dataset		MDX23 Leaderboard
Algorithm name	SDR Drums	SDR Other	SDR Drums
HTDemucs4	12.04	16.56	---
MelBand Roformer	12.76	17.28	---
SCNet Large	13.01	17.53	---
SCNet XL	13.42	18.00
MelBand + SCNet XL Ensemble	13.78	18.31	---
BS Roformer SW	14.11	---	---
MelBand + SCNet XL + BS Roformer SW Ensemble	14.35	---	---

Detailed statistics on Multisong dataset:

Model	Drums fullness	Drums bleedless	Drums SDR	Drums L1Freq	Other fullness	Other bleedless	Other SDR	Other L1Freq
HTDemucs4	15.36	25.00	12.04	37.47	33.03	37.22	16.56	38.37
MelBand Roformer	14.16	42.12	12.76	40.80	33.97	47.24	17.28	42.02
SCNet Large	14.91	28.23	13.01	38.04	35.39	35.03	17.53	39.36
SCNet XL	21.21	24.47	13.42	40.30	38.56	38.32	18.00	40.35
MelBand + SCNet XL Ensemble	19.66	30.23	13.78	41.74	38.09	42.90	18.31	42.00
BS Roformer SW	14.78	43.70	14.11	42.23	---	---	---	---
MelBand + SCNet XL + BS Roformer SW Ensemble	16.97	39.73	14.35	42.74	---	---	---	---

MVSep Drums (drums, other)

Advanced features

Company

Extra