An automatic music transcription based on translation of spectrum and sound path estimation

Abstract

An automatic music transcription method is proposed. The method is based on a generative model that takes into account the translation of spectrum for an instrument and the sound path from the instrument to a microphone. The fundamental frequency (note), the spectrum of the instrument (basis pattern) and the sound path are estimated simultaneously using an extended complex nonnegative matrix factorization. The effectiveness of the proposed method is confirmed by synthetic data. © 2011 Springer-Verlag.