Iterative Audio Separation with Mixture Consistency via MIMO Model Extension
arXiv - 2026
Pinned Sony README: MUSDB18-HQ, museval median-of-medians. Compare only within this protocol, separately from MVSEP and other paper scores.
Loading...
Separate lead vocals, harmonies and accompaniment, or extract dry vocals for covers, karaoke and vocal editing.
Input
Supports inputs in formats such as mp3 / wav / flac / ogg / m4a / aac / aif / aiff / mp4 / mov / mkv
Supported formats
Output
2 × WAV
Use case
MelBand Aspiration separates breath and aspiration from other sounds, with standard and less aggressive checkpoints. Similar sounds such as fricatives may also be extracted; use isolated vocals and check articulation.
Choose a model for your material and purpose, and review the examples before submitting.
Cost, time, formats and parameters depend on the selected model.
Include the tool name, steps and the problem in your feedback. Screenshots can help us locate it.
MelBand Aspiration separates breath and aspiration from other sounds, with standard and less aggressive checkpoints. Similar sounds such as fricatives may also be extracted; use isolated vocals and check articulation.
Sony MIMO BS-RoFormer separates vocals and accompaniment, with Large and Small versions. Complex mixes may still contain bleed.
Sony MIMO BS-RoFormer separates vocals and accompaniment, with Large and Small versions. Complex mixes may still contain bleed.
Runtime wrapper for UVR/MVSep-style separator checkpoints. It provides the model loading, WAV stem export, and FFmpeg-backed audio format handling used by this tool.
MVSep 9205 supplies the lead/accompaniment-with-harmony leg, BS-RoFormer Leap XE 90 bands supplies vocals with harmony, and BS-RoFormer Leap Instrumental supplies the accompaniment reference.
Leap XE 90 bands supplies the lead vocal, BS-RoFormer Leap Instrumental supplies the accompaniment reference, MVSep 9205 BS-Kar AVG removes backing harmonies, and the anvuew BS-RoFormer dereverb checkpoint strips reverb.
arXiv - 2026
Pinned Sony README: MUSDB18-HQ, museval median-of-medians. Compare only within this protocol, separately from MVSEP and other paper scores.
SDX23 music separation track - 2023
BS-RoFormer uses band-split spectrogram features plus RoPE Transformers; the SDX23 system ranked first in the music source separation track and a smaller version reported 9.80 dB average SDR on MUSDB18HQ.
TISMIR - 2024
The SDX23 music demixing paper documents robust MSS benchmarks, listening tests, and that the best system improved over the prior challenge by more than 1.6 dB SDR on MDXDB21.
arXiv / ISMIR late-breaking demo - 2023
Mel-RoFormer replaces heuristic non-overlapping bands with overlapping mel-band subbands and reports better separation than BS-RoFormer for vocals, drums, and other stems on MUSDB18HQ.
2 output categories
Only records with verified public inference implementations and matching weights are shown; algorithms requiring no weights are also eligible. Closed services, unreleased weights and unmatched experiment versions are excluded. This is not an all-model leaderboard.
Each set of weights appears once per benchmark, using its best published result for the selected metric. An overall score comes from one evaluation. Expand to compare evaluations and settings.
Public weights do not establish a fully open-source license. Expand the data for declared weight licenses and access conditions. Undeclared licenses are marked separately.
Markers identify models or components used here. Scores and settings come from the cited tests, not a new evaluation of the complete site pipeline.
No comparable public results with verified code and matching weights are available.
Scores apply to the listed dataset, metric and model version; results from different tests are not directly comparable.
These values come from released filenames without a verified common dataset or full evaluation protocol. SDR and L1Freq are not comparable and do not establish a quality ranking.
Public sources checked: 2026-10-01
MelBand Aspiration
Aspiration · checkpoint metadataMelBand Aspiration Less Aggressive
Aspiration · checkpoint metadata