Source Separation and Beat Tracking: A System Approach to the Development of a Robust Audio-to-Score System

摘要

A set of tools for pitch, metric and rhythm information matching on musical audio streams are under development. The goal is a robust automatic system capable to track musical audio data for information retrieval and access purpose. Several problems have been identified and approached to be solved separately, such as metric computation, rhythm recognition and pitch tracking. This approach has been chosen to synthesize a whole processing model robust enough to be applied virtually to any kind of music. A source separation modeling has been investigated starting from ICA model. An ICA modified unmixing model has been proposed as preprocessing subsystem. This subsystem demonstrated to be very helpful for efficient applying of processing algorithms to rhythm and pitch tracking.