A Music Identification System Based on Chroma Indexing and Statistical Modeling

A methodology is described for the automatic identification of classical music works. It can be considered an extension of fingerprinting techniques because the identification is carried out also when the query is a different performance of the work stored in the database, possibly played by different instruments and with background noise. The proposed methodology integrates an already existing approach based on hidden Markov models with an additional component that aims at improving scalability. The general idea is to carry out a clustering of the collection to highlight a limited number of candidates to be used for the HMM-based identification. Clustering is computed using the chroma features of the music works, hashed in a single value and retrieved using a bag of terms approach. Evaluation results are provided to show the validity of the combined approaches.