Section: New Results

Semi-Markov Models for Real-time MIDI-to-Score Alignment

We develop a new stochastic model of symbolic (MIDI) performance of polyphonic scores, based on Semi-Markov models, to align MIDI performances of music scores. In our approach, the evolution of the music performer and the production of performed notes are modeled with a hierarchical extension of hidden semi-Markov models (HSMM). By comparing with a previously studied model based on hidden Markov model (HMM), we give theoretical reasons why the present model is advantageous to deal with complex music event such as trills, tremolos, arpeggios, and other ornaments. This is also confirmed empirically by comparing the accuracy of score following and analysing the errors. We also develop a hybrid of this HSMM-based model and the HMM-based model which is computationally more efficient and retains the advantages of the former model. The present model yields one of the state-of-the-art score following algorithms for symbolic performance and can possibly be applicable for other music recognition problems. Details and results are published in [19] .

This work was done in collaboration with Eita Nakamura from the National Institute of Informatics of Tokyo, Japan.