stems-models / README.md
0xfff999's picture
Model card: float16 Mega 53 core
d541780 verified
|
Raw History Blame Contribute Delete
1.52 kB
---
license: other
tags:
- audio-source-separation
- onnx
---
# Model files for the stems website
Files served to the browser by https://github.com/federava/stems. Each keeps its own licence:
| Files | What | Licence | Authors |
|---|---|---|---|
| `ggml-model-htdemucs*.bin`, `onnx/htdemucs*.onnx` | Demucs v4 (HT Demucs), converted | MIT | Meta Platforms (Alexandre Défossez et al.), https://github.com/adefossez/demucs |
| `onnx/idm-analyse.onnx`, `onnx/idm-synth.onnx`, `onnx/idm-mel.bin` | Inverse Drum Machine (drum kit split), exported | Apache-2.0 | Bernardo Torres, Geoffroy Peeters, Gaël Richard, https://github.com/bernardo-torres/inverse-drum-machine |
| `onnx/mel-roformer-kim.onnx` | Mel-Band RoFormer vocals, ONNX export by audiojs (unchanged copy of phuongnse/mel-roformer-kim-onnx) | MIT | Kimberley Jensen, https://huggingface.co/KimberleyJSN/melbandroformer |
| `onnx/mega53-core.onnx`, `onnx/mega53-core-f16.onnx`, `onnx/mega53-<instrument>.onnx` | MVSep Mega 53-stem BS-RoFormer, split into a shared core and one head per instrument (exported, float16 weights; the `-f16` core also computes in float16) | MIT | Roman Solovyev (ZFTurbo, MVSep), https://github.com/ZFTurbo/Music-Source-Separation-Training (release v1.0.21; licence statement in issue #245) |
| `onnx/basic-pitch.onnx` | Basic Pitch (audio to notes), unchanged | Apache-2.0 | Spotify AB, https://github.com/spotify/basic-pitch |
Full credits and licence texts: https://github.com/federava/stems/blob/main/THIRD_PARTY_NOTICES.md