Add HT-Demucs ONNX (fp16 weights), re-hosted from StemSplitio@d54ed9eb with MIT notice and attribution

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) Meta Platforms, Inc. and affiliates.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,114 @@
+---
+license: mit
+pipeline_tag: audio-to-audio
+tags:
+  - music-source-separation
+  - stem-separation
+  - demucs
+  - htdemucs
+  - onnx
+  - onnxruntime-web
+library_name: onnx
+---
+
+# htdemucs-onnx (fp16 weights)
+
+An ONNX export of **HT-Demucs** (Hybrid Transformer Demucs, `htdemucs`) for
+4-stem music source separation — drums, bass, other, vocals.
+
+This repository is a **re-host, not a new model**. It exists so that
+[Audio Magic](https://audiomagic.studio) has a stable, pinned copy of a
+browser-loadable Demucs export rather than depending on a third-party repo that
+could move or change. Nothing has been modified.
+
+## Provenance
+
+| | |
+|---|---|
+| Original model | [Demucs v4 / HT-Demucs](https://github.com/adefossez/demucs) — Meta Platforms, MIT |
+| ONNX conversion | [`StemSplitio/htdemucs-onnx`](https://huggingface.co/StemSplitio/htdemucs-onnx), revision `d54ed9eb60e258ea82131c6ee14578628816456a` |
+| File | `htdemucs_fp16weights.onnx` → `htdemucs_fp16.onnx` |
+| Size | 165,612,636 bytes |
+| SHA-256 | `d05c269d0178d2a72ad484b10b11dd370193fc923201c3b27a99f848745db70a` |
+
+The file is byte-identical to the upstream conversion at that revision. Verify with:
+
+```bash
+shasum -a 256 htdemucs_fp16.onnx
+```
+
+## Inputs and outputs
+
+| | Name | Shape | Dtype |
+|---|---|---|---|
+| Input | `mix` | `(1, 2, 343980)` | float32 |
+| Output | `stems` | `(1, 4, 2, 343980)` | float32 |
+
+- Stereo, 44.1 kHz, 7.8-second segments (343,980 samples), samples in `[-1, 1]`.
+- Output stem order: **`[drums, bass, other, vocals]`**.
+- For longer audio, process in overlapping segments and overlap-add.
+
+## Verification
+
+Checked with `onnxruntime-node` on CPU against a synthetic four-source mixture
+with known ground truth:
+
+- **Reconstruction: 37.0 dB SNR** — the four stems sum back to the input, the
+  property Demucs is built to have, which is the strongest available evidence
+  that the conversion is faithful.
+- **Drums** correlate 0.99 and **bass** 1.00 with their ground-truth sources,
+  with negligible cross-leakage.
+- Vocals were not meaningfully testable this way: a synthetic harmonic tone is
+  not singing, and the model reasonably classified it as an instrument. Vocal
+  quality should be judged on real recordings.
+- Speed: 1.1 s per 7.8 s segment on native CPU — a realtime factor of ~0.14×.
+
+## Licence
+
+Demucs and its pretrained weights are released by Meta under the MIT licence,
+which permits redistribution provided the copyright notice travels with it. The
+full notice is included in this repository as `LICENSE` and reproduced here:
+
+```
+MIT License
+
+Copyright (c) Meta Platforms, Inc. and affiliates.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+```
+
+## Citation
+
+If you use this, cite the original authors — the work is theirs:
+
+```bibtex
+@inproceedings{rouard2023hybrid,
+  title     = {Hybrid Transformers for Music Source Separation},
+  author    = {Rouard, Simon and Massa, Francisco and D{\'e}fossez, Alexandre},
+  booktitle = {ICASSP 23},
+  year      = {2023}
+}
+
+@inproceedings{defossez2021hybrid,
+  title     = {Hybrid Spectrogram and Waveform Source Separation},
+  author    = {D{\'e}fossez, Alexandre},
+  booktitle = {Proceedings of the ISMIR 2021 Workshop on Music Source Separation},
+  year      = {2021}
+}
+```

htdemucs_fp16.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d05c269d0178d2a72ad484b10b11dd370193fc923201c3b27a99f848745db70a
+size 165612636