Note Recognition of Polyphonic Music by Using Timbre Similarity and Direction Proximity
Yohei Sakuraba, Hiorshi G. Okuno · University of Michigan Library Repository · 2003
Note recognition in automatic music transcription consists of two processes: simultaneous grouping and sequential grouping. The former generates a note from frequency components, while the latter generates a temporal sequence of notes. Their main problem are disambiguation in note composition and design of features effective for music stream creation, respectively. For the simultaneous grouping, to cope with the problem note hypotheses are created based on overlap detection of frequency components. For the sequential grouping, timbre similarity and direction proximity are integrated. The result of experiments with quartet music recorded in an anechoic chamber showed that the proposed method improved the F-measure of each grouping by 0.10 and 0.14, respectively.