2002 Fiscal Year Final Research Report Summary
Musical Information Processing by using Sound Ontology
| Project/Area Number |
12480090
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Intelligent informatics
|
|---|
| Research Institution | KYOTO UNIVERSITY (2001-2002)
Tokyo University of Science (2000) |
|---|
Principal Investigator |
OKUNO Hiroshi Graduate School of Informatics, Professor, 情報学研究科, 教授 (60318201)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
KAWAHARA Tatsuya Graduate School of Informatics, Associate Professor, 情報学研究科, 助教授 (00234104)
GOTO Masataka National Institute of Advanced Industrial Science and Technology, Scientist, 情報処理研究部門, 研究員
|
|---|
| Project Period (FY) |
2000 – 2002
|
| Keywords | Sound Ontology / Musical Instrument Identification / F0-dependent Multivariate Normal Distribution / Localization by Intramural Phase Difference / Hierarchical Recognition of Musical Instruments / Unsupervised Learning / Clustering / Decision Tree Learning |
|---|
| Research Abstract |
(1) Musical instrument identification by using pitchdependency
F0-dependent multivariate normal distribution is invented to represent the F0 (pitch) dependency of acoustical features of musical instruments so that F0-dependency and F0-independency are represented by the mean and variance of the F0-dependent multivariate normal distribution. The resulting performance of musical instrument identification is improved by 16.48% at individual instrument level and by 20.6% at category level
(2) Musical sound source separation by combining timbre similarity and localization
The ambiguities of polyphony music caused by overlapping frequency components are resolved by using sound source localization. Based on this schema, musical instrument identification system of polyphony music has been developed, which first extracts mono-tones by simultaneous grouping and then constitutes part streams by sequential grouping. The effectiveness of the proposed system was proved by using quartet performance recorded in an anechoic room
(3) Percussion instrument identification
The drums and cymbals are discriminated by low-pass and high-pass filters. The former is identified by unsupervised learning to absorb the differences of individual instruments and performance, while the ambiguities of cymbals and snare drums are resolved by automatic learning of heuristics concerning their error patterns
(4) Systematic construction of sound ontology
From the features represented by F0-dependent multivariate normal distribution, the hierarchical representation of musical instruments has been created. Thus the method of systematic generation of sound ontology is established. In addition, new observation concerning the pitch-dependency of musical instrument hierarchy has been obtained
|
