著者
安部 翔 小田 弘良 松島 俊明
雑誌
研究報告音楽情報科学(MUS)
巻号頁・発行日
vol.2011, no.1, pp.1-6, 2011-10-04

我々は自動譜めくりシステムの開発を進めており,そのために多重音の音高・音源数の高速推定の研究を行っている.以前,複素スペクトル内挿法により基本周波数候補を求め,それらの組み合わせによる同時発音数を評価する尺度を導入し,この評価尺度が最小なる音高の組み合わせを直接求める方法で,音源数が未知の音響信号からの高速推定法について報告したが,検出可能な同時発音数を多くするに従って計算量も増え,処理落ちが時々生じるという問題点があった.そこで,今回,この評価尺度の最小値の変化率の推移に着目し,これを用いて多重音の音高・音源数の高速推定を試みた結果,以前よりも高速な推定が可能となったので報告する.We are developing an automatic page turner system, therefore are studying high-speed estimation of the number of notes and their pitches simultaneously from acoustic sound. We already reported the following method: that is, to calculate the candidates of fundamental frequencies depending on the complex-spectrum interpolation, to introduce a evaluation function to search for the combination of the pitches which gives the minimum value of the evaluation function directly, and to presume the number of notes and their pitches from acoustic sound to a high speed. However, by this method, computational complexity also increased as the number of detective notes increased, and there was a problem that processing omission sometimes occurred. By considering transition of the rate change of the minimum value of the evaluation function, faster presumption of the number of notes and their pitches from acoustic sound can be attained than before.