Computer Science Colloquium
Eita Nakamura
Kyoto University and Centre for Digital Music at Queen Mary University of London
Statistical Performance Model with Explicit Voice Structure and Symbolic Music Alignment
Wed 06.09.2017, 13:45, 90 minutesRoom JKU S3-055 in Science Park 3
Abstract
The central problem in symbolic music alignment (a task of matching musical notes in performance MIDI files to the corresponding notes in the musical score) is how to deal with the varieties in human performances (e.g. tempo changes, performance errors, ornaments, repeats and skips, etc.). To solve the problem, the approach of applying generative statistical models of music performance is discussed. In particular, to deal with possible asynchrony between two hands in piano performances, the merged-output hidden Markov model is proposed as a model that explicitly incorporates the multiple-voice structure of polyphonic music. As applications, automatic music accompaniment and an alignment system are demonstrated.Bio
Eita Nakamura is a JSPS Postdoctoral Research Fellow in the Speech and Audio Processing Group at Kyoto University and he is currently an academic visitor at the Centre for Digital Music at Queen Mary University of London. He received a PhD in Physics at the University of Tokyo in 2012 and have published papers on various topics in symbolic music processing including automatic music accompaniment, music transcription, and automatic music arrangement. His research interests include music modelling and analysis, music information processing, and statistical machine learning.Invited by Univ.-Prof. Dr. Gerhard Widmer, Department of Computational Perception
The Computer Science Colloquium is organized by the Department of Coputer Science at JKU, the Österreichische Gesellschaft für Informatik (ÖGI) and the Österreichische Computergesellschaft (OCG).