A JAX/Flax implementation of audio processing functions, inspired by and building upon librosa and TorchLibrosa.

Although, it is optional, we recommend first installing the jax-ai-stack with one of these three options:

pip install jax-ai-stack              # JAX CPU
pip install jax-ai-stack "jax[cuda]"  # JAX + AI stack with GPU/CUDA support
pip install jax-ai-stack "jax[tpu]"   # JAX + AI stack with TPU support

Required: Then install librosax:

pip install librosax

Documentation is here.

This library is heavily inspired by and borrows code from:

• librosa - The excellent Python library for audio and music analysis by the librosa development team
• TorchLibrosa - PyTorch implementations of librosa functions and neural net layers by Qiuqiang Kong
• nnAudio - PyTorch implementations of CQT and other functions by Kin Wai Cheuk

librosax is licensed under the ISC License, matching the license used by librosa. See the LICENSE file for details.