Who is Vocal / accompaniment separation (RoFormer) for?

Separation & stem split Download vocal and accompaniment WAV stems

Vocal / accompaniment separation (RoFormer) · TelkNet

Separation & stem splitGPU workload

Vocal / accompaniment separation

Q: What tasks can Vocal / accompaniment separation (RoFormer) help with?

Separation & stem split; Download vocal and accompaniment WAV stems; Positioning: open-weight, locally deployable two-stem vocal/accompaniment separation for practice, covers, mixing, and sampling prep.; Full songs, cover sources, or mixed audio; BS-RoFormer Leap XE 90 bands outputs vocals and BS PolarFormer public ONNX 62 bands outputs accompaniment.; Vocal / accompaniment separation: extract vocal material; Vocal / accompaniment separation: export accompaniment material

Q: Which model and keyword entities are connected with Vocal / accompaniment separation (RoFormer)?

BS-RoFormer Leap XE 90 bands (vocals output), BS PolarFormer public ONNX 62 bands (accompaniment output), MVSep 10178 Leap XE, pcunwa Leap XE weights, BS-RoFormer, MVSep 10009 PolarFormer public, BS PolarFormer public ONNX, audio-separator runtime

Positioning: open-weight, locally deployable two-stem vocal/accompaniment separation for practice, covers, mixing, and sampling prep.

Download vocal and accompaniment WAV stems

4 credits/useFile inputMax 50 MBEstimated 180-900 seconds

Input

Full songs, cover sources, or mixed audio; BS-RoFormer Leap XE 90 bands outputs vocals and BS PolarFormer public ONNX 62 bands outputs accompaniment.

Supported formats

mp3wavflacoggm4aaacaifaiffmp4movmkv

Output

Download vocal and accompaniment WAV stems

Use case

Positioning: open-weight, locally deployable two-stem vocal/accompaniment separation for practice, covers, mixing, and sampling prep.

Models, parameters, and sources

Open-weight vocal/accompaniment separation: the vocal stem comes from BS-RoFormer Leap XE 90 bands, and the accompaniment stem comes from BS PolarFormer public ONNX 62 bands for editing, covers, transcription prep, and downstream audio work.

BS-RoFormer Leap XE 90 bands (vocals output)BS PolarFormer public ONNX 62 bands (accompaniment output)MVSep 10178 Leap XEpcunwa Leap XE weightsBS-RoFormerMVSep 10009 PolarFormer publicBS PolarFormer public ONNXaudio-separator runtimepython-audio-separatoraudio-separator PyPIaudio-separator model weightsMVSep metrics + verified weights

BS-RoFormer Leap XE 90 bands (vocals output)

MVSep 10178 Leap XE pcunwa Leap XE weights BS-RoFormer

BS PolarFormer public ONNX 62 bands (accompaniment output)

MVSep 10009 PolarFormer public BS PolarFormer public ONNX BS-RoFormer

audio-separator runtime

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.

python-audio-separator audio-separator PyPI audio-separator model weights

Public leaderboard: Verified public-weight vocals ranking / Verified public-weight accompaniment ranking

Open-source deployable scope: these tables include only rows with verified downloadable weights or configurations and a local deployment path. SOTA and Best are relative to the verified rows shown, not hosted/API-only or commercial systems and not a universal leaderboard.

Verified public-weight vocals ranking

MVSep metrics + verified weights

Verified public-weight Multisong vocals SDR rows only. TelkNet uses Leap XE 90 for vocals; hosted, API-only, or unverifiable MVSep rows are excluded from this ranking.

Rank	Model	Vocals SDR (dB)	Detailed metrics
MVSep entry 10178	BS-RoFormer Leap XE 90 bands (pcunwa)TelkNetMVSep 10178 metric	11.7577SOTA	Instrumental SDR (dB)17.5303
MVSep entry 10009	BS PolarFormer public ONNX 62 bands (bgkb/ZFTurbo)MVSep 10009 metric	11.7575	Instrumental SDR (dB)18.0650Best
MVSep entry 10173	BS-RoFormer Leap vocal-only (pcunwa)MVSep 10173 metric	11.7398
MVSep entry 10146	BS-RoFormer Leap 62 bands (pcunwa)MVSep 10146 metric	11.7222	Instrumental SDR (dB)17.7794

Verified public-weight accompaniment ranking

MVSep metrics + verified weights

Verified public-weight Multisong instrumental SDR rows only. TelkNet uses BS PolarFormer public ONNX 62 for accompaniment; hosted, API-only, or unverifiable MVSep rows are excluded.

Rank	Model	Instrumental SDR (dB)	Detailed metrics
MVSep entry 10009	BS PolarFormer public ONNX 62 bands (bgkb/ZFTurbo)TelkNetMVSep 10009 metric	18.0650SOTA	Vocals SDR (dB)11.7575
MVSep entry 10146	BS-RoFormer Leap 62 bands (pcunwa)MVSep 10146 metric	17.7794	Vocals SDR (dB)11.7222
MVSep entry 10178	BS-RoFormer Leap XE 90 bands (pcunwa)MVSep 10178 metric	17.5303	Vocals SDR (dB)11.7577Best

Paper and benchmark notes

Music Source Separation with Band-Split RoPE Transformer

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.

arXiv:2309.02612

The Sound Demixing Challenge 2023 - Music Demixing Track

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.

TISMIR 2024

Official / repositories

MVSep 10178 Leap XE pcunwa Leap XE weights MVSep 10009 PolarFormer public BS PolarFormer public ONNX python-audio-separator audio-separator PyPI audio-separator model weights

Papers / technical notes

BS-RoFormer

Parameter guide

Separate first: split vocals/accompaniment or stems before transcribing the target audio.
Save separated tracks: keeps stems or track outputs for separate download.
Dedicated route: this tool uses the model path selected for this workflow.

Use cases

Vocal / accompaniment separation: extract vocal material
Vocal / accompaniment separation: export accompaniment material
Vocal / accompaniment separation: prepare cleaner stems for MIDI or RVC

Workflow

1Upload the full song or cover source
2Run the Leap XE 90 vocals leg and the PolarFormer public 62 accompaniment leg separately, then export two real stems.
3Download vocal and accompaniment WAV stems

Pre-flight checklist

Prefer clear audio with limited clipping, distortion, or reverb
Vocal / accompaniment separation: check vocal and accompaniment separately
Vocal / accompaniment separation: processing material, not final mastering repair