# ▶️ umx.js - web demo

Web Audio Paper (PDF) (opens new window) Repository (opens new window)

In this demo, we showcase an audio separation models that performs separation in the browser. The model is derived from the full open-unmix. The models were converted to tensorflow via ONNX and on-the-wire weight quantization was applied to reduce the models file size. We manually converted the python-based pre-and post-processing pipelines from Python to JavaScript. Thus, as part of the release of this demo, we make a fully invertible tensorflow.js pipeline available.


Note, that the model is very slow and we do not advice to separate tracks longer than 30 seconds.

# Cite

  title={open.unmix.app-towards audio separation on the edge},
  conference="2021 Web Audio Conference",
      St{\"o}ter, Fabian-Robert and Machry, Maria Clara and de Andrade Vaz, Delton and Uhlich, Stefan and Mitsufuji, Yuki and Liutkus, Antoine}