Librosa

Librosa is a Python package for music and audio analysis, providing the building blocks necessary to create music information retrieval (MIR) systems. Developed by Brian McFee, Eric Battenberg, and others at UC San Diego and LabROSA (Columbia University), Librosa has over 7,000 stars as of 2026 and is the most widely used audio analysis library in Python's scientific ecosystem. The library provides a comprehensive suite of audio processing functions organized into submodules: core (audio loading via soundfile/audioread, resampling, time-frequency representations including Short-Time Fourier Transform, Constant-Q Transform, and Inverse STFT for spectrogram reconstruction), feature (spectral feature extraction: chroma features (chroma_stft, chroma_cqt, chroma_cens), mel-scaled spectrogram and MFCC (Mel-Frequency Cepstral Coefficients with configurable n_mfcc, typically 13-40), spectral centroid, spectral bandwidth, spectral contrast, spectral flatness, spectral rolloff, zero crossing rate, poly features, tonnetz, RMS energy), rhythm (tempo estimation via onset strength envelope, beat tracking with dynamic programming, tempo histogram, plp pulse curve), onset (onset detection via spectral flux with multiple onset detection functions, onset strength envelope, onset backtracking), beat and tempo, decompose (non-negative matrix factorization for source separation, harmonic-percussive source separation via median filtering), effects (time stretching via phase vocoder, pitch shifting via time stretch + resample, harmonic extraction, percussive extraction, trim, split, preemphasis), display (spectrogram visualization with matplotlib, chroma and tonnetz plots, waveforms with color-coded segments), and segments (structural segmentation via self-similarity matrix, agglomerative clustering, cross-similarity). Librosa integrates with NumPy, SciPy, SoundFile, Numba (JIT compilation), and scikit-learn.