Hi awesome members,
Looking for solutions, particularly any tool that can split up recorded voice such as soprano, bass, alto, and tena separately from the mixed record.
A free open source tool will be much appreciated.
Vocal removers pull the choir OUT of music — wrong tool.
You need to split choir voices FROM each other: soprano / alto / tenor / bass.
Free tools exist for this exact problem. All below.
🌐 Browser-only — SATB separation, zero install
MVSEP — upload file, pick model, download split stems:
https://mvsep.com/en/algorithms
→ Start with #104 (multi-stem vocal): https://mvsep.com/algorithms/104
HuggingFace Spaces — free GPU, runs in browser:
- All models in one UI: https://huggingface.co/spaces/r3gm/Audio_separator
- UVR5 with BS-Roformer + MDX23C: https://huggingface.co/spaces/TheStinger/UVR5_UI
- SOTA Mel-RoFormer + HTDemucs: https://huggingface.co/spaces/JacobLinCool/vocal-separation
- ZFTurbo MDX23: https://huggingface.co/spaces/Ma5onic/MVSEP-MDX23-music-separation-model
🖥️ UVR5 + MSST-WebUI — free desktop apps, offline, no coding
UVR5 (Ultimate Vocal Remover):
https://github.com/anjok07/ultimatevocalremovergui
→ Use MDX-Net or BS-Roformer models inside — not the default Spleeter (= old, not built for choir)
→ Choir threads: https://github.com/Anjok07/ultimatevocalremovergui/discussions/800
MSST-WebUI (browser UI, all latest models):
https://github.com/SUC-DriverOld/MSST-WebUI
→ One-click install: https://github.com/xavikla555/MSST-Installer
→ Free Colab run: https://github.com/SUC-DriverOld/MSST-WebUI/blob/main/webUI_for_colab.ipynb
PolUVR (UVR with Gradio UI, GPU):
https://github.com/Bebra777228/PolUVR
🔬 SATB-specific tools — built for choir, not hacked from vocal removers
UMSS — inputs mixed choir → outputs 4 voice streams (S/A/T/B):
- Audio demo: https://schufo.github.io/umss/
- Code: https://github.com/schufo/umss
- ICASSP 2024 (newer): https://github.com/PierreChouteau/umss_icassp | https://pierrechouteau.github.io/umss_icassp
Choral Music Separation (ISMIR 2022):
- Demo: https://retrocirce.github.io/cms_demo/
- Code: https://github.com/RetroCirce/Choral_Music_Separation
MTG Barcelona (top-3 music tech lab) — full SATB toolkit:
- Choral voice extraction: https://github.com/MTG/content_choral_separation
- Voice assignment in SATB quartets: https://github.com/helenacuesta/voas-vocal-quartets
- SATB separation + pitch extraction: https://github.com/pc2752/choir_separation_f0_analysis
- Live demos: https://helenacuesta.github.io/demos.html
Pretrained Conditioned-U-Net — hear SATB results directly:
http://www.dariuspetermann.com/satb-source-separation-results/
SepACap (2025, a cappella multi-singer):
- Code: https://github.com/ETH-DISCO/SepACap
- Paper + demo: https://arxiv.org/abs/2509.26580
🧲 Choir-trained models — load into UVR5 or MSST, skip generic ones
Chorus Male+Female BS-Roformer (splits male/female first → then refine per voice):
https://huggingface.co/Sucial/Chorus_Male_Female_BS_Roformer
MSST full model pack with choral models included:
https://huggingface.co/Sucial/MSST-WebUI
MelBand Roformer 4-stem fine-tuned large:
https://huggingface.co/SYH99999/MelBandRoformer4StemFTLarge
All public UVR models collected:
https://huggingface.co/Blane187/all_public_uvr_models
ZFTurbo model zoo + community model drops:
https://github.com/ZFTurbo/Music-Source-Separation-Training/blob/main/docs/pretrained_models.md
https://github.com/ZFTurbo/Music-Source-Separation-Training/issues/1
🎼 Stem → sheet music — convert separated voice audio to notation
Basic-Pitch (Spotify) — audio to MIDI, single voice, free browser tool:
https://basicpitch.spotify.com/
Omnizart — full transcription including singing voice:
https://github.com/Music-and-Culture-Technology-Lab/omnizart
Colab: https://colab.research.google.com/github/Music-and-Culture-Technology-Lab/omnizart/blob/master/colab.ipynb
💡 5 situations where choir separation actually pays off
- Choir director → extract each voice part separately so singers can practice their part in headphones with everyone else muted
- Music teacher → isolate one student’s voice section from a group recording to check if they’re singing the right notes
- A cappella group → old choral recording has no sheet music — separate each part, run through Basic-Pitch, get notation
- Sound engineer → client’s choir recording has one voice part too loud — separate, re-balance, re-mix cleanly
- Music student → famous choir has a beautiful harmony you want to learn — pull each voice apart and transcribe it yourself
Vocals were never locked. Just mixed. Now you have the unmixer.
!