Merged-Output Hidden Markov Model and Its Applications in Music Information Processing

2014 Fiscal Year Final Research Report

• Project/Area Number: 25880029
• Research Category: Grant-in-Aid for Research Activity Start-up
• Allocation Type: Single-year Grants
• Research Field: Intelligent robotics
• Research Institution: Meiji University (2014); National Institute of Informatics (2013)
• Principal Investigator: NAKAMURA Eita 明治大学, 研究・知財戦略機構, 研究推進員(ポスト・ドクター) (10707574)
• Research Collaborator: SAGAYAMA Shigeki 明治大学, 総合数理学部, 教授 (00303321) ; ONO Nobutaka 国立情報学研究所, 情報学プリンシプル研究系, 准教授 (80334259) ; WATANABE Kenji 東京藝術大学, 音楽学部, 教授
• Project Period (FY): 2013-08-30 – 2015-03-31
• Keywords: 出力合流隠れマルコフモデル / 統計的音楽モデル / 統計的演奏モデル / 自動伴奏 / 自動採譜 / ピアノ運指 / 自動編曲

Outline of Final Research Achievements

A statistical model that describes music phenomena with multiple musical instruments and voice parts, named merged-output hidden Markov model (HMM), is constructed and applied to music information processing. The model is constructed by merging the outputs from multiple HMMs, each of which corresponds to one instrument or one voice part. Efficient inference algorithms for the model are derived. The model is applied to music processing tasks including automatic music accompaniment, music transcription, piano fingering optimisation, voice part separation, and automatic piano arrangement, yielding processing techniques with higher performance than conventional methods.

Free Research Field

音楽情報処理