Audio Playground models (Demucs v4, ONNX)

Browser-ready versions of Meta's Hybrid Transformer Demucs, used by Rabiat's Audio Playground. The app runs them on the visitor's GPU with ONNX Runtime Web, so no audio ever leaves the device.

File Stems Size
htdemucs.onnx vocals, drums, bass, other 86 MB
htdemucs_6s.onnx vocals, drums, bass, other, guitar, piano 57 MB

Only the neural network is exported; the STFT and inverse STFT run in JavaScript. Weights are stored as float16 and cast to float32 inside the graph. Output matches PyTorch within 54-57 dB.

Original models: facebookresearch/demucs, MIT license, © Meta Platforms, Inc.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support