Granular Feature-Space Synthesizer

This project explores an interactive approach to granular synthesis by representing audio grains in a perceptually informed feature space. Audio files are segmented into short grains, and each grain is analyzed using spectral and perceptual descriptors including spectral centroid, spectral flatness, pitch confidence, and loudness.

The resulting feature dataset is organized using a k-d tree to enable efficient k-nearest-neighbor queries, allowing real-time, controllable clustering and traversal of similar sounds. An interactive interface visualizes each grain as a point in feature space, enabling continuous navigation, playback, and dynamic effects processing, and supporting intuitive exploration of timbre and texture in real time.