pyspch: speech analysis and recognition package for teaching

What is it ?

pyspch is a Python package that provides easy to use speech analysis and speech recognition functionality in Python.

The focus is on EASE of USE and SIMPLE ACCESS to the science and algorithms behind speech processing today. This is by no means intended to be a state-of-the-art package for building speech recognition systems.

The principle usage of pyspch is with Jupyter Notebooks in which small systems or demonstrators are built, such as the suite of demonstrations and exercises in the spchlab repository( https://github.com/compi1234/spchlab)

Main features

• Jupyter Notebook toolkit for speech visualization and audio I/O
• speech analysis:
  • feature extraction: MEL and CEP
  • postprocessing and grouping
  • spectrograms (Fourier, MEL and CEP)
• speech recognition:
  • dynamic time warping
  • Bayesian classifiers
  • Hidden Markov Models
  • Deep Neural Nets (Multilayer Perceptrons)

Installation

The source code resides on GitHub at: https://github.com/compi1234/pyspch

pip install git+https://github.com/compi1234/pyspch.git

Dependencies

• recommended Python versions are 3.10 or 3.9
• torch and torchaudio are recommended, but not required for most of the package except the nn subpackage (DNNs)
  • installing with conda on Windows (do this after creating your normal env)

    conda install pytorch==2.1.0 torchaudio==2.1.0 pytorch-cuda=11.8 -c pytorch -c nvidia

  • You may find a more compatible version for your system on the pytorch website