pyAudioAnalysis: Analyze, Classify and Segment Audio

Overview

pyAudioAnalysis is a Python library for audio feature extraction, classification, segmentation and applications, developed by MagCIL at NCSR Demokritos. It gives Python users and command-line workflows a single toolkit for turning audio files into features and applying machine-learning models to classify, segment or otherwise analyze audio.

The library covers both supervised tasks, such as training a classifier on labeled audio folders, and unsupervised analysis, including speaker diarization. Its README links to a wiki for detailed usage guidance and examples.

Key Features

• Extract audio representations including MFCCs, spectrograms and chromagrams.
• Train, tune and evaluate classifiers for audio segments, then classify new sounds.
• Detect audio events and remove silence from long recordings.
• Perform supervised segmentation that combines segmentation and classification.
• Perform unsupervised segmentation, including speaker diarization, and extract audio thumbnails.
• Train and apply audio regression models, including for emotion-recognition applications.
• Reduce dimensionality to visualize audio data and similarities.
• Run analysis tasks through Python wrappers or command-line commands.

Use Cases

• A researcher can extract audio features and evaluate segment classifiers for a labeled sound dataset.
• A speech or media team can use unsupervised segmentation to explore recordings and identify speaker turns.
• A developer can detect events or exclude silence when processing long audio recordings.
• A machine-learning practitioner can visualize audio representations or experiment with regression for an application such as emotion recognition.

Getting Started

git clone https://github.com/tyiannak/pyAudioAnalysis.git
pip install -r ./requirements.txt
pip install -e .

See the README and project wiki for examples and task-specific guidance.

Considerations

• The project is marked maintained and has had a few commits recently, but it has published no release in over a year.
• Most of the code comes from one person, despite a wider contributor set. No issues were closed and no pull requests were merged in the recent measured period, so do not assume fast community support.
• The supplied requirements identify Python and tests; they do not specify a required GPU, database or container runtime.
• The project uses the Apache-2.0 license. Review its terms for your distribution and use case.