Dictionary-based pursuit decomposes any sound into grains, enabling analysis-driven transformations

Dictionary-based pursuit (DBP), also known as the matching-pursuit family of algorithms, is the analytical counterpart to granular synthesis. It seeks by iterative search to match the energy in a signal with a vast dictionary of millions of sound atoms or grains, proceeding from the strongest time-frequency unit to the weakest. Because Gabor showed the granular paradigm applies to the analysis of any sound, DBP can decompose an arbitrary recording, not just a synthetic one. The result is a highly malleable time-frequency representation of grains that opens transformations such as pluriphonic spatial scattering of a sound’s particles or ‘cavitation’ processes that carve new granular patterns out of existing ones.

Examples

MPTK (Matching Pursuit Toolkit) for analysis; Roads’s Scatter app for artistic manipulation of DBP output — a Wigner-Ville display of per-grain energy that can be filtered, dragged, or scrubbed.

Assessment

What is dictionary-based pursuit and how does it relate to granular synthesis? Describe one transformation that becomes possible once a sound is decomposed into grains via DBP.