A System for the Automatic Identification of Music Works

This paper describes a system able to identify a music work through the analysis of the audio recording of a performance. The approach is based on the statistical modeling of the expected audio features of music performances, given a database of known music works. In particular, the automatic identification is based on an application of hidden Markov models (HHMs), which are automatically built from music scores available in digital format. States of the HMMs are labeled by score events, and transition and observation probabilities are directly computed from the information on the score. Three alternative approaches to the identification task have been proposed and tested on a set of audio excerpts. Results showed that the methodology can achieve satisfactory results. A prototype system has been developed, and will be demonstrated, which allows in a few seconds to identify an unknown recording from a dataset of hundreds of scores.