audioFlux

Why we included this project

Most of the work in audio and music machine learning happens before a model ever sees the data, and audioFlux is built to take over that front end. It wraps dozens of time-frequency transforms, from STFT and CQT to wavelet variants and non-stationary Gabor, along with several hundred feature combinations covering spectral features, cepstral coefficients, chroma, and pitch estimators like YIN. Those features plug directly into deep-learning training for classification, source separation, and music information retrieval, so it reads as a practical feature-extraction layer rather than a research toy. The core is written in C, which keeps large dataset runs and even mobile use fast. Developers prototyping on Linux, macOS, or Windows will find the Python API easy to adopt, with documented examples spanning MFCC extraction, onset detection, and harmonic-percussive separation.