The pretrained models for Open-Unmix, mirrored from Zenodo, which serves them too slowly for a Space to start. Each folder is one model with one checkpoint per target, under the filenames openunmix expects.
umxl/: UMX-L, licensed CC-BY-NC-SA-4.0, so results cannot be used commercially and must be shared alike.umxhq/: UMX-HQ, licensed MIT.
The card declares CC-BY-NC-SA-4.0, the more restrictive of the two. The third published model, UMX, is left out: it is UMX-HQ trained on audio band-limited to 16 kHz, kept upstream for comparison with older results.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support