asteroid-team/asteroid

Health v2 · maintenance only

Adoption not part of the score

2584 stars · 450 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Asteroid is a PyTorch-based audio source separation toolkit for researchers, providing modular building blocks (filterbanks, encoders, maskers, decoders, losses), datasets, and recipes to reproduce papers. It also ships pretrained models for speech separation and enhancement.

Use cases

• separate overlapping speakers in an audio recording
• enhance noisy speech with deep learning
• reproduce a source separation paper's results
• experiment with new separation architectures in PyTorch
• download pretrained speech separation models
• train a model on the WSJ0-2mix dataset

When to choose

• you research audio or speech source separation and want modular, extensible PyTorch building blocks
• you need reproducible recipes covering data prep, training, and evaluation
• you want pretrained separation/enhancement models out of the box

When to avoid

• you need a production-ready end-user app rather than a research toolkit
• your task is general music remixing with stems from a service rather than research experimentation
• you don't use PyTorch or Python

Facets

library · maturity active

machine-learning deep-learning audio-processing speech-recognition machine-learning deep-learning speech-processing python cross-platform pytorch source-separation speech-enhancement pretrained-models research audio gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="asteroid-team/asteroid")