作者: Bryan Pardo , William P. Birmingham
DOI: 10.1162/014892602760137167
关键词: Chord (music) 、 Segmentation 、 Algorithm 、 Computer science 、 Generality 、 Music theory 、 Multiple time dimensions 、 Computer music 、 Parsing 、 Harmony (Music)
摘要: Computer Music Journal, 26:2, pp. 27–49, Summer 2002 O Massachusetts Institute of Technology. The harmonic analysis tonal music by computer is an important area interest in the research community. While problem interesting its own right, ability to parse and use chords harmonies a composition also adds dimension agent’s manipulate musical material. Automated arranging, accompaniment, phrasing are all aided understanding chordal structure piece. Although musicians analyze types, it dif cult create algorithms that have this generality. In article, we describe set for make minimal stylistic contextual cues, such as metric strength, context, known constraints. system described article purposefully simple approach. This allows us specify workings point where work may be duplicated extended others. designed both explore basic computational properties solve task provide baseline against which other systems can measured. way, possible measure change performance introduced things voiceleading rules, metrical information. We divide into two tasks. Segmentation splitting appropriate chunks (segments) analysis. As default, assume segmentation takes place time dimension. Labeling giving each segment proper quality root pitch. Figure 1 shows partitioned labeled segments. perform labeling. Previous automated (Winograd 1968; Smoliar 1980; Laden Keefe 1989; Ebcioglu 1992; Maxwell Widmer Smaill et al. 1993; Taube 1999), with notable exception recent Temperley Sleator (1999), has either avoided issue taking already segmented input, or been unclear about how done. clear ef cient do address problem, they only return name (rather than chord quality) not give any statistical their when compared outside measure. To our knowledge, no researchers published results on music. lack measures literature makes comparisons among cult. present empirical evaluation using corpus 45 excerpts compiled David Temperley, from Kostka Payne theory textbook (Kostka 1984), hereafter referred KP corpus. (The appendix lists piece used corpus.) analyses provided answer key based teacher’s edition grading later article. Both individual provided, classes error tabulated reported.