Vast musical databases are currently accessible over computer networks (e.g., the Web), creating a need for sophisticated methods to search and organize these databases. Because music is a multifaceted, multi-dimensional medium, it demands specialized representations, abstractions and processing techniques for effective search that are fundamentally different from those used for other retrieval tasks. By exploiting reductionist theories of musical structure and performance (i.e., musical style), this project will develop hierarchical, stochastic music representations and concomitant storage and retrieval mechanisms that are well-suited to music's unique characteristics, and are both musically and psycho-acoustically plausible. A software system exploiting these representations and retrieval mechanisms will be developed that accepts sonic input, compares abstractions of this input to those in a database of digital recordings, returns sonic samples of the database that best match the query, and allows the user to refine the query using music/acoustic-based interfaces of varying degrees of complexity. This research will yield a working music-search engine will application to e-commerce and will provide new scientific knowledge in terms of algorithms and mathematic models in the fields of databases, information retrieval, artificial intelligence, signal processing and perception, for music search and retrieval, which will generalize to other multimedia processing tasks.
