---
title: BandBuddy-Models
canonical_url: "https://www.modelscope.cn/models/Zzzzzzorz/BandBuddy-Models"
md_url: "https://www.modelscope.cn/models/Zzzzzzorz/BandBuddy-Models.md"
repository: Zzzzzzorz/BandBuddy-Models
chinese_name: "BandBuddy Audio Separation Assets"
last_updated: 2026-09-11
license: GPL-3.0
pipeline_tag: speech-separation
tasks:
  - speech-separation
library_name:
  - onnx
downloads: 0
stars: 0
tags:
  - audio-source-separation
  - music-source-separation
  - bandbuddy
---

# BandBuddy-Models

> BandBuddy-Models - Zzzzzzorz 在 ModelScope 开源的模型。BandBuddy 2.0 音频分轨运行资源

Zzzzzzorz/BandBuddy-Models 是 ModelScope 魔搭社区上的speech-separation模型，采用 GPL-3.0 许可。

- **Repository**: Zzzzzzorz/BandBuddy-Models
- **License**: GPL-3.0
- **Tasks**: speech-separation
- **Tags**: audio-source-separation, music-source-separation, bandbuddy
- **Downloads**: 0
- **Stars**: 0
- **Last updated**: 2026-09-11

Source: https://www.modelscope.cn/models/Zzzzzzorz/BandBuddy-Models

---

# BandBuddy Audio Separation Assets

BandBuddy 2.0.1 downloads the six files below from the immutable `v2.0.1` release tag. The existing `v2.0.0` tag remains available for older applications.

High and balanced quality use a shared acoustic/electric BS-RoFormer plus the Lead/Rhythm Mel-Band RoFormer. Fast quality uses acoustic and electric MDX-Net models plus the Lead/Rhythm HTDemucs model.

## Verified files

| Path | Bytes | SHA-256 |
|---|---:|---|
| `5c90dfd2-34c22ccb.th` | 54996327 | `34c22ccb381c6f9fdbf324f04e1e2fe21aaaf293f5ded163a162697ff9a02ddd` |
| `bs_mega_53stem_acoustic-electric_shared_mvsep.ckpt` | 102410137 | `183607bffbebdb43dcb3fd583b7cbe3c77fb55aea886cbd1e3aa316d9148698c` |
| `mbr_lead_rhythm_guitar_listra92.ckpt` | 337073664 | `b3c47bca33609ca1ba0bb2d2076410bfd1eb941b051b72afc1f3e24d12b17eef` |
| `mdx_6s_acoustic_guitar_anvuew.onnx` | 27147460 | `2bd8f2af629b279cc1a568f895ee9636f7ce2d76c69aa601e6744eaab8b4916a` |
| `mdx_6s_electric_guitar_anvuew.onnx` | 27147623 | `bd6fcf40659771568ee180ea69bd4576a9c3d2423ae0f5f6f5afcc8b6a6fd938` |
| `demucs4_lead_rhythm_guitar_drypaint.ckpt` | 109822623 | `946ffd50d7f2fd87e447d880525283e88bd9061e1b428d4f1d380e764d54d618` |

The shared checkpoint was rebuilt from the two verified original Mega 53-stem checkpoints using BandBuddy's `scripts/build-v2.0.1-model-bundle.py`. Every shared tensor is identical between the sources; both original mask heads are preserved without retraining. The saved checkpoint was reloaded and all tensors checked against those sources. `build-audit.json` records the result. This published binary's checksum is pinned in the application.

## Provenance

- Original acoustic/electric: `noblebarkrr/BS-Roformer-MVSep-Mega-53-stems`, revision `0677941f9cdfd891a6bc3336229244e197035327`.
- Lead/Rhythm and fast models: `noblebarkrr/mvsepless_resources`, revision `46ca65d52c07635a4992fadb9c6b1644eb51fe7f`.
- Inference architecture: Music-Source-Separation-Training commit `0e5f1159fc5ea87fc13b957584e178b4977e5dd3`.

See `manifest.json` for quality profiles and configuration hashes, and `THIRD_PARTY_NOTICES.md` for attribution.