librosa

Librosa is a Python library for music and audio analysis. It provides tools for loading audio, extracting features, and building analysis pipelines. It emphasizes simple, consistent interfaces that integrate with NumPy and SciPy.

Core capabilities include audio input/output (for example, librosa.load), sampling rate handling and resampling, and time-frequency analysis such as the short-time Fourier transform and the corresponding amplitude and power spectrograms. It offers several spectral representations, including Mel spectrograms, MFCCs, chroma features, as well as measures like spectral centroid, bandwidth, rolloff, and zero-crossing rate. Librosa also includes harmonic-percussive source separation (HPSS) for foreground-background separation.

Feature extraction and analysis features cover tempo estimation and beat tracking, onset detection, and pitch tracking using Yin and PyIN. It supports time-stretching and pitch-shifting and provides higher-level representations such as constant-Q transform (CQT) and chroma-based analyses, enabling a range of music information retrieval workflows.

Applications and ecosystem: Librosa supports visualization through the librosa.display module and is commonly used in teaching, research, and prototyping of audio-analysis pipelines. It is designed to work well with NumPy/SciPy and relies on optional dependencies such as audioread for reading certain file formats.

Availability and impact: Librosa is an open-source Python package with active development and broad user adoption in the music information retrieval and signal-processing communities. Documentation and the project repository provide tutorials, API references, and example workflows.