Instructions to use moonbox-dev/separation with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use moonbox-dev/separation with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download moonbox-dev/separation --local-dir separation
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Download README.md from moonbox-dev/separation: direct link, hf CLI and curl.
- Browser
- Download file 2.79 kB
-
https://huggingface.co/moonbox-dev/separation/resolve/main/README.md
- Command line
-
hf download hf://moonbox-dev/separation/README.md
-
curl -L -o README.md https://huggingface.co/moonbox-dev/separation/resolve/main/README.md
license: mit
library_name: mlx
pipeline_tag: audio-to-audio
tags:
- mlx
- moonbox
- speech-enhancement
- music-source-separation
DeepFilterNet3 and BS-RoFormer for Moonbox
The files the Moonbox desktop app downloads for these models:
- DeepFilterNet3 (Hendrik Schröter): Takes background noise out of voice recordings, at four strengths.
- BS-RoFormer (ZFTurbo and MVSep): Splits a song into four stems, or into the instruments you pick from 53.
They are the original checkpoints converted to MLX for Apple silicon. Moonbox checks every file against a signed list before using it. You do not need this repository to use Moonbox, and it is not an official release of the original models.
What is here
| Path | What | From |
|---|---|---|
models/mlx/df3.* |
DeepFilterNet3 | https://github.com/Rikorose/DeepFilterNet (models/DeepFilterNet3.zip) |
models/mlx/bsr4.* |
BS-RoFormer 4-stem (ZFTurbo, model_bs_roformer_ep_17_sdr_9.6568.ckpt, MSST v1.0.12) | https://github.com/ZFTurbo/Music-Source-Separation-Training/releases/tag/v1.0.12 |
models/mlx/bsr53.* |
MVSep Mega 53-stem BS-RoFormer (mvsep_mega_model_bs_roformer_53_stems_v1.ckpt, MSST v1.0.21) | https://github.com/ZFTurbo/Music-Source-Separation-Training/releases/download/v1.0.21/mvsep_mega_model_bs_roformer_53_stems_v1.ckpt |
Every part was converted, quantized or repackaged from its source unless its row says "as published": treat none of these files as the original release.
Licences
The 4-stem BS-RoFormer was trained by its author on MUSDB18-HQ, a dataset released for non-commercial research; the 53-stem model's training data is not disclosed by its authors. The weights themselves are MIT-licensed; whether the training data matters for your use is yours to judge.
Credit for the stem models: ZFTurbo (Music-Source-Separation-Training) and MVSep.
| Part | Licence | Notice |
|---|---|---|
| DeepFilterNet3 | MIT License; Apache License 2.0 | Copyright 2021 Hendrik Schröter. Dual-licensed: MIT or Apache-2.0. |
| BS-RoFormer 4-stem (ZFTurbo, model_bs_roformer_ep_17_sdr_9.6568.ckpt, MSST v1.0.12) | MIT License | Copyright (c) 2024 Roman Solovyev (ZFTurbo). Credit: ZFTurbo. |
| MVSep Mega 53-stem BS-RoFormer (mvsep_mega_model_bs_roformer_53_stems_v1.ckpt, MSST v1.0.21) | MIT License | Copyright (c) 2024 Roman Solovyev (ZFTurbo). Credit: ZFTurbo and MVSep. |
The full text of each licence is in LICENSE. The models are the work of their authors, linked above. Moonbox converted and packaged them and claims no rights in them. Use is subject to each licence above.