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
Upload the full song or cover source
Supported formats
Output
Download harmony without accompaniment, vocals with harmony, harmony with accompaniment, accompaniment without harmony, and vocals without harmony
Use case
Five-stem harmony split with harmony vocals, vocals with harmony, harmony accompaniment, clean accompaniment, and lead-vocal cleanup for covers, practice, and detailed mixing
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.
MVSep 9205 BS-Kar AVG public-weight chain runs on the full mix; Leap XE and BS-RoFormer Leap Instrumental supply the reference stems.
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.
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.
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.
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.
5 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.
Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.
The overall view uses lead-vocal SDR, the output measured for every model in this board. Isolated backing vocals and backing vocals with accompaniment are separate categories and are not averaged together.
| Metric rank | Model | Lead vocal↑SDR (dB) ↑ | Harmony vocals↑SDR (dB) ↑ | Accompaniment↑SDR (dB) ↑ | Instrumental + backing vocals↑SDR (dB) ↑ |
|---|---|---|---|---|---|
| 1 | BS-Kar-Gabox_IS + BS-Kar-Frazer&Becruily + BS-Kar-Anvuew (AVG)Pipeline componentPublic weights · Inference codeBS-Kar-Gabox_ISBS-Kar-Frazer&BecruilyBS-Kar-AnvuewInference code bsroformers/bs_karaoke_gabox_IS.ckptEvaluation settingsWeight license not declared | 10.4296 | - | - | 15.8159 |
| 2 | bs_karaoke_anvuew.ckptPublic weights · Inference code | 10.1586 | - | - | 15.5449 |
| 3 | bs_karaoke_becruily.ckptPublic weights · Inference code | 10.1005 | - | - | 15.4868 |
| 4 | bs_karaoke_gabox.ckptPublic weights · Inference code | 10.0246 | - | - | 15.4109 |
| 5 | mbr_karaoke1_gilliaaan.ckptPublic weights · Inference codePublic weightsInference code mel_band_roformer/mbr_karaoke1_gilliaaan.ckptWeight license not declared | 9.8160 | - | - | 14.8532 |
| 6 | bs_pope_karaoke_974_lambda.ckptPublic weights · Inference code | 9.7257 | - | - | 15.1120 |
| 7 | mbr_karaoke_small_gabox_aufr33.ckptPublic weights · Inference codePublic weightsInference code mel_band_roformer/mbr_karaoke_small_gabox_aufr33.ckptWeight license not declared | 9.6911 | - | - | 15.0774 |
| 8 | mbr_karaoke2_gabox.ckptPublic weights · Inference code | 9.6822 | - | - | 15.0684 |
| 9 | mbr_karaoke1_gabox.ckptPublic weights · Inference code | 9.6798 | - | - | 15.0661 |
| 10 | mbr_karaoke_becruily.ckptPublic weights · Inference code | 9.6148 | - | - | 14.2185 |
| 11 | bs_pope_karaoke2_lambda.ckptPublic weights · Inference code | 9.4050 | - | - | 14.7913 |
| 12 | mbr_karaoke_aufr33_viperx.ckptPublic weights · Inference codePublic weightsInference code mel_band_roformer/mbr_karaoke_aufr33_viperx.ckptWeight license not declared | 9.3260 | - | - | 14.7123 |
| 13 | mbr_karaoke28022025_gabox.ckptPublic weights · Inference codePublic weightsInference code mel_band_roformer/mbr_karaoke28022025_gabox.ckptWeight license not declared | 9.2269 | - | - | 14.6132 |
| 14 | mbr_karaoke25022025_gabox.ckptPublic weights · Inference codePublic weightsInference code mel_band_roformer/mbr_karaoke25022025_gabox.ckptWeight license not declared | 8.6502 | - | - | 14.0365 |
| 15 | bs_karaoke_3stem_giantailab.ckptPublic weights · Inference code | 8.5410 | 4.1583 | 14.2295 | - |
| 16 | bs_karaoke_inv_gabox.ckptPublic weights · Inference code | 8.3846 | - | - | 13.7722 |
| 17 | mbr_karaoke_fusion_aggr_gonzaluigi.ckptPublic weights · Inference codePublic weightsInference code mel_band_roformer/mbr_karaoke_fusion_aggr_gonzaluigi.ckptWeight license not declared | 8.3818 | - | - | 13.7681 |
| 18 | mbr_karaoke_fusion_gonzaluigi.ckptPublic weights · Inference codePublic weightsInference code mel_band_roformer/mbr_karaoke_fusion_gonzaluigi.ckptWeight license not declared | 8.3583 | - | - | 13.7446 |
| 19 | mbr_karaoke_fusion2_aggr_gonzaluigi.ckptPublic weights · Inference codePublic weightsInference code mel_band_roformer/mbr_karaoke_fusion2_aggr_gonzaluigi.ckptWeight license not declared | 6.6540 | - | - | 12.0403 |
| 20 | 5_hp-karaoke-uvr.ckptPublic weights · Inference code | 6.5766 | - | - | 11.9430 |
| 21 | mdx_karaoke1.onnxPublic weights · Inference code | 6.4123 | - | - | 11.7986 |
| 22 | 6_hp-karaoke-uvr.ckptPublic weights · Inference code | 6.3516 | - | - | 11.7202 |
| 23 | mbr_karaoke_fusion_total_aggr_gonzaluigi.ckptPublic weights · Inference codePublic weightsInference code mel_band_roformer/mbr_karaoke_fusion_total_aggr_gonzaluigi.ckptWeight license not declared | 6.2167 | - | - | 11.6030 |
| 24 | mgm-v5-karokee-32000-beta1.ckptPublic weights · Inference code | 5.7597 | - | - | 11.1547 |
| 25 | mgm-v5-karokee-32000-beta2-agr.ckptPublic weights · Inference code | 5.6449 | - | - | 11.0187 |
| 26 | mdx_karaoke2.onnxPublic weights · Inference code | 4.7226 | - | - | 10.1088 |
Markers identify models or components used here. Scores and settings come from the cited tests, not a new evaluation of the complete site pipeline.
Datasets, versions and conditions remain separate, including sources with no eligible ranking rows.
Deployable entries 26 / source rows 127 · Checked: 2026-10-25
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
Mel-RoFormer (Karaoke / Duet) · Ensemble of 5 Karaoke Models (2026.10) · MVSep Lead/Back BS Roformer 124 bands (2026.10) Overlap: 8 · MVSep Lead/Back BS Roformer 124 bands (2026.10) · BS-Kar-Gabox_IS + BS-Kar-Frazer&Becruily + BS-Kar-Anvuew + BS-Kar-ByMVSepTeaм (AVG) · MVSep Lead/Back BSRoformer (2025.09) · deux karaoke · BSkarfrazerBecruily+BSkarMVSEP+MBkarGaboxV2 · BS-karaoke-Frazer&Becruily + BS-Karaoke-Anvuew (AVG) · BS-Kar-FrazerBecruily+MB-Kar-Becruily+BS-Kar-byMVSepTeam · BSFrazerBecruily + MBBecruily · BS Roformer Karaoke by anvuew · karaoke gabox · BS-Kar-Gabox_IS + BS-Kar-Frazer&Becruily + BS-Kar-Anvuew (Max Spec) · BS-Roformer Karaoke anvuew · sami-bytedance · KarGaboxV2+KarAufr33ViperX+KarFrazerBecruily +KarBecruily Average/Average · BS-Roformer Karaoke (frazer & becruily) · BS Roformer Karaoke Gabox · gilliaan mbrofo karaoke V1 · MVSep Karaoke MelBand Roformer (fused model gabox + aufr33/viperx) · gilliaan_melroformer_karaoke_v1 · karaoke_gaboxv2 · karaoke gabox v1 · Ensemble (MelBand Roformer aufr33 and becruily) · BS PolarFormer SoloVoice TEST29 · BSPolafFormer SoloVoice Ensemble V1+V2 · MelBand_Karaoke_BVE_Gonzaluigi · Small Karaoke Gabox&Aufr · gabox_frazer&becruily_anvuew fused model · roformer-harmoney-5 · SCNet BVE · SCNet BVE + MelRoformer BVE · MelBand Roformer Karaoke by becruily · BSPFSV TEST 42 · SCNet BVE + MelRoformer BVE + Becruily Karaoke · last_bs_roformer by gabox · Potato · BS-RoFormer Karaoke · BSPFSV TEST 40 · BS PolarFormer SoloVoice TEST39 · BSPFSV TEST 41 · roformer-harmoney-2 · BSPolafFormer SoloVoice TEST№2 · BSPolafFormer SoloVoice TEST№4 · BSPolafFormer SoloVoice TEST№3 · roformer-harmoney-3 · BSPolafFormer SoloVoice TEST · MelBand Roformer Karaoke by aufr33 и viperx · Fuzed BS-kar-Frazer\Becruily + BS-kar-Anvuew 0.5/0.5 · BSPFSV TEST 43 · MelRoformer BVE · test_lead_back_0907 · MelBand Roformer Karaoke by aufr33 и viperx (extract from vocals) · MelBand Roformer Karaoke by becruily (extract from vocals) · 111521 · Diff-VS (Karaoke / Duet) · roformer_harmoney · tf-yingmusic-3-stem · BS PolarFormer Karaoke TEST15 · Karaoke 7 · gilliaan_bsroformer_deeffect_v1beta · DeMIX Pro (extract from vocals) · MDX-B (Karaoke, extract from vocals) · Karaoke 8 · Mel Band RoFormer Duet · DeMIX Pro · BS PolarFormer Karaoke TEST6 · BS PolarFormer Karaoke TEST5 · MDX-B (Karaoke) · BS PolarFormer Karaoke TEST7 · Karaoke6 · BS PolarFormer Karaoke TEST10 · MelBand Karaoke Fusion Total by Gonzaluigi Nyan · BS PolarFormer Karaoke TEST11 · BS PolarFormer Karaoke TEST8 · BS PolarFormer Karaoke TEST13 · BS PolarFormer Karaoke TEST9 · BS PolarFormer Karaoke TEST14 · BS PolarFormer Karaoke TEST12 · HTDemucs - wesleyr36 · UVR Karaoke model (karokee_4band_v2_sn) · BVEv1+BVEv2Aufr33 Avg · UVR Karaoke model (HP-KAROKEE-MSB2-3BAND-3090) · Karaoke 5 · BS PolarFormer Karaoke TEST4 · BS PolarFormer Karaoke TEST3 · BS PolarFormer Karaoke TEST2 · Karaoke 4 · Kimberley Jensen MelBand Roformer (Vocal model) · BS PolarFormer Karaoke TEST1 · MelBand Roformer by becruily · MelBand Roformer by becruily (moddifed) · MelBand Roformer Karaoke by Gabox (moddifed) · MelBand Roformer Karaoke by Gabox · UVR Back Vocals model (UVR-BVE-4B_SN-44100-1) · becruily karaoke high fullness
Find processing methods and steps suited to your material.