What is Asteroid?

Asteroid is a PyTorch-based audio source separation toolkit.

The main goals of Asteroid are:

• Gather a wider community around audio source separation by lowering the barriers to entry.

• Promote reproducibility by replicating important research papers.

• Automatize most engineering and make way for research.

• Simplify model sharing to reduce compute costs and carbon footprint.

So, how do we do that? We aim to provide

• PyTorch Dataset for common datasets.

• Ready-to-use state-of-the art source separation architectures in native PyTorch.

• Configurable recipes from data preparation to evaluation.

• Pretrained models for a wide variety of tasks and architectures.

Who is it for?

Asteroid has several target usage:

• Use asteroid in your own code, as a package.

• Use available recipes to build your own separation model.

• Use pretrained models to process your files.

• Hit the ground running with your research ideas!

Want to know more?

• Visit our webpage

• Read our paper

• Watch the presentation video

• Check how we won the PyTorch Hackathon 2020 !