Analysis, Recognition, Manipulation and Generation of Music Signal and Information based on Mathematical Models

点击翻译按钮获取中文摘要

An diverse research area has been actively investigated in analysis, recognition, manipulation and generation of music signal and information resulting in numerous excellent results. They include quite a wide range of research achievements such as multipitch estimation method (called HTC) based on a model of multiple concurrent music sounds and another method based on non-negative matrix factorization (NMF), harmonic-percussive sound separation (HPSS), high-quality signal manipulation (e.g., music tempo and pitch modification) based on restoration of phase components from power spectrogram, part separation form stereophonic music recordings, vocal sound extraction and deletion from ensemble music signals based on spectral fluctuation of human voice, chord estimation from music signals, automatic extraction of constituent rhythms (RhythmMap) in music signals and bar-line segmentation, music genre and mood recognition based on the previous method, rhythm estimation in polyphonic music signals and application to automatic music transcription, music modeling and automatic transcription based on dynamic Bayesian network to integrate hierarchical layers from music theory to signal observations, composer identification from score data, automatic music performance rendition using conditional random field, automatic counterpoint with imitation of music motives, automatic song composition based on intrinsic intonation of Japanese lyrics, and building functional harmony database. These achievements have been made public through journal papers, international conferences, review papers, domestic academic presentations, web service for automatic music composition, and public media.