A Learning-Based Quantization: Estimation of Onset Times in a Musical Score

摘要

This paper describes a method for organizing onset times of musical notes performed along a jam-session accompaniment into the normalized (quantized) positions in a score. The purpose of this study is to align onset times of a session recording to quantized positions so the performance data can be stored in a reusable form. Unlike most previous beat-tracking-related methods focusing on predicting or estimating beat positions, our method deals with the problem of eliminating the onset-time deviations under the condition that the beat positions are given. To quantize polyphonic MIDI recordings of jam session, we propose a method that uses hidden Markov models for modeling onset-time transition and deviation. Its main advantage is that a player's performance is quantized using a model learned statistically from session recordings of that player. Experimental results show that our model performs better than the semi-automatic quantization in commercial sequencing software.