Open-Unmix UMX-HQ other (ONNX)
An ONNX export of the other network from Open-Unmix
UMX-HQ (pretrained weights: Zenodo 10.5281/zenodo.3370489,
other-b52fbbf7.pth, MIT). The Mix Directive desktop app uses it, with the
vocals, drums, bass and other networks
together, for Music Rebalance (re-levelling everything that is not voice, drums or bass) and Reduce Bleed's Model mode. It runs locally with ONNX Runtime on the CPU. No PyTorch is required.
Changes from the original
This is a modified redistribution under the original MIT licence (see LICENSE, © 2019 Inria):
- Only the other target's network is included, converted to ONNX (opset 17).
- The graph is traced through a thin wrapper that repeats
OpenUnmix.forwardwith a dynamic frame axis. The original reads the frame count as a Python int, which tracing would freeze. Batch size is fixed at 1 and stereo input is fixed. - The STFT/iSTFT and Wiener filtering are not part of the graph. The app turns the four targets' outputs into power-ratio masks instead.
The weights are unchanged (float32). The ONNX output matches PyTorch to within 4e-06.
The newer umxl weights are licensed for non-commercial use only and are not used.
Files
| File | Size | SHA-256 |
|---|---|---|
model.onnx | 35,623,507 bytes | fe62c1d8fb8326c7e68540a675dfa186a2b131f763fb41050ea5e9a9859eb76e |
export_umxhq_onnx.py, export_umxhq_vocals_onnx.py | — | Scripts that produced model.onnx (python3 export_umxhq_onnx.py out_dir other) |
Usage
- Input
magnitude: float32[1, 2, 2049, frames]. This is the magnitude STFT of a 44.1 kHz stereo signal (n_fft 4096, hop 1024, periodic Hann window, centred with reflect padding, as intorch.stft). - Output
other_magnitude: the estimated other magnitude, the same shape as the input. Use it with the other targets' estimates as power-ratio masks on the complex mix STFT, then invert. - The network is a bidirectional LSTM, so give it a few seconds of context either side when processing in chunks.
Training data
The upstream authors trained UMX-HQ on MUSDB18-HQ and publish the weights under the MIT licence. See the Open-Unmix repository and the Zenodo record.
Citation
Stöter, F.-R., Uhlich, S., Liutkus, A., & Mitsufuji, Y. (2019). Open-Unmix – A Reference Implementation for Music Source Separation. Journal of Open Source Software, 4(41), 1667. https://doi.org/10.21105/joss.01667