HTDemucs (ONNX) — copy used by Pizu Audio
Unmodified copy of the HTDemucs music source separation model (Meta) in ONNX format, used by Pizu Audio to split songs into vocals, drums, bass and other — entirely in the user's browser (ONNX Runtime Web, WebGPU / WebAssembly). No audio is uploaded anywhere.
| File | htdemucs_embedded.onnx |
| Size | 180 534 758 bytes |
| SHA-256 | e5e425c17683f163a472462eb5f5a4ffcd11c31858d57fbd0833b012d8b88077 |
| Input | 44.1 kHz stereo, segments of 343 980 samples (see demucs-web) |
| Outputs | drums, bass, other, vocals |
Origin and license
- Model and weights: HTDemucs by Meta Platforms, Inc. —
facebookresearch/demucs, MIT License
(full text in
LICENSE). - ONNX export: demucs-web by timcsy (MIT),
originally published at timcsy/demucs-web-onnx
(revision
92e33df61cfc9eb820272aaa62d2ef6dcf4d950d).
This repository only mirrors the file so that Pizu Audio does not depend on a third-party account. The file is byte-identical to the original (same SHA-256).