audlib

A speech signal processing library in Python with emphasis on deep learning.

audlib provides a collection of utilities for developing speech-related applications using both signal processing and deep learning. The package offers the following high-level features:

• Speech signal processing utilities with ready-to-use applications
• Deep learning architectures for speech processing tasks in PyTorch
• PyTorch-compatible interface (similar to torchvision) for batch processing
• A command-line interface with a unix-pipe-like syntax

Some use cases of audlib are:

• Extracting common speech features for your backend
• Developing your own deep-learning-based tools for speech tasks
• Quickly try out speech processors and visualize the spectrogram in command line

audlib focuses on correctness, efficiency, and simplicity. Signal processing functionalities are mathematically checked whenever possible (e.g. constant overlap-add, istft(stft(X))==X). Deep neural networks follow the PyTorch's convention.

Breaking Changes

• 0.0.3
  • sig.util.freqz becomes sig.util.fftfreqz
  • nn.strfnet is removed. See the official repository here.
  • transform.stlogm is removed
• 0.0.2
  • audioread follows the interface of soundfile.read
  • audiowrite follows the interface of soundfile.write
  • The argument sr is removed from all short-time transforms

Installation

pip install audlib

Developer Installation

In the source directory, install the library with test dependencies:

pip install ".[tests]"

Run test:

python -m pytest tests

Release flow

1. Bump version in setup.py.
2. Package release: python setup.py sdist bdist_wheel
3. Upload release: twine upload --repository-url https://upload.pypi.org/legacy/ dist/*

Usage example

More extensive examples can be found in examples/.

Release history

• 0.0.3
  • First release of the command-line tool audpipe
• 0.0.2
  • Streamlines optional installation
  • Improves API (see breaking changes)
  • Adds coverage test
• 0.0.1
  • First release on PyPI

Contact

Please reach out to Raymond Xia (raymondxia@cmu.edu) for comments.

More information about our research groups can be found on at

• CMU Robust Speech Recognition Group
• Machine Learning for Signal Processing Group

License

Distributed under the MIT license. See LICENSE for more information.