任意箇所への弾き直し・弾き飛ばしを含む演奏に追従可能な楽譜追跡と自動伴奏

Bibliographic Information

Other Title
  • Score Following Handling Performances with Arbitrary Repeats and Skips and Automatic Accompaniment

Search this article

Abstract

人間の演奏に対して,楽譜に基づき,自動的に同期して伴奏することは自動伴奏と呼ばれる.これは,合奏曲の練習や生演奏と電子音楽の融合に応用可能な技術であり,その効用は大きい.特に,練習時においては,弾き誤りや弾き直し,弾き飛ばしを含む演奏が起こりうるが,従来の自動伴奏はそのような演奏に対して追従が困難であるか,または特定箇所への弾き飛ばしのみ対応している.本稿では,任意箇所への弾き飛ばしを含む演奏に対しても素早く追従可能な自動伴奏を議論する.まず,隠れマルコフモデルを用いて演奏の生成過程をモデル化し,その確率的逆問題を解く形で,楽譜追跡アルゴリズムを設計する.そして,この際に起こる処理時間増大の問題を解決するために,最尤状態推定の高速アルゴリズムであるαβγ法を考案する.この楽譜追跡手法の有効性を確認するために,実際の演奏を用いて評価を行う.さらに,これをもとに自動伴奏システムEurydiceを構築し動作を確認する.

Automated performance of an accompaniment score, synchronized to human performances, is called automatic accompaniment. It is applicable for practicings of ensemble music and live electronics, and its utility is large. Particularly in music practicings, performances with errors, repeats and skips are expected. However, conventional automatic accompaniment systems have difficulties in following those performances, or may handle only skips to specific locations in scores. In this paper, we study a robust automatic accompaniment capable of following performances with skips to arbitrary locations. We construct a hidden Markov model which describes the generation process of performances and propose a score following algorithm, which solves the probabilistic inverse problem of the model. To solve the problem of increasing processing time, a fast algorithm for estimating most likely states is proposed. To confirm the validity of the proposed score following algorithm, an evaluation test for the algorithm with human-played performances is performed. Furthermore, we construct an automatic accompaniment system Eurydice and confirm its operation.

Journal

Related Projects

See more

Details

Report a problem

Back to top