Musifind: A System for the Automatic Identification of Music Works

This paper describes a working music retrieval prototype, based on a methodology for the recognition of audio recordings. The recognition is based on an application of statistical models, which are automatically built from digital music scores. States of the HMMs are labeled by score events, and the transition and observation probabilities are directly computed from the information on the score. Six approaches to the recognition task have been tested on a set of audio excerpts. Tests showed that the methodology can achieve satisfactory results, at least for a supervised labeling of audio recordings.