Asteroid is a PyTorch-based audio source separation toolkit that enables fast experimentation on common datasets and on a wide range of deep neural network architectures.

More information about the toolkit can be found on our github repository.

You can also read our paper (published in INTERSPEECH 2020) available here.

We are looking forward to transferring our technologies by enhancing what has already been released in Asteroid with new features. Including real time processing and multichannel processing and also adapting these methods to your use cases.

For more information, feel free to contact us :

• Antoine Deleforge, supervisor : [email protected]
• Manuel Pariente, creator of Asteroid : [email protected]
• Joris Cosentino, engineer working on Asteroid: [email protected]

This page shows selected audio examples of speech separation and speech enhancement results obtained with Asteroid.

You can check out these results by clicking on the top left menu selector.