Mixtures of probability experts for audio retrieval and indexing

Author

Slaney, Malcolm

Author_Institution

IBM Almaden Res. Center, San Jose, CA, USA

Volume

1

fYear

2002

fDate

2002

Firstpage

345

Abstract

This paper describes a system for connecting nonspeech sounds and words using linked multidimensional vector spaces. An approach based on a mixture of experts learns the mapping between one space and the other. This paper describes the conversion of audio and semantic data into their respective vector spaces. Two different mixture-of-probability-expert models are trained to learn the association between acoustic queries and the corresponding semantic explanation, and vice versa. Test results are presented based on commercial sound effects CD.

Keywords

audio databases; content-based retrieval; data mining; database indexing; expert systems; learning (artificial intelligence); probability; acoustic queries; audio indexing; audio retrieval; commercial sound effects CD; linked multidimensional vector spaces; mapping learning; mixture-of-probability-expert models; nonspeech sounds; semantic data; semantic explanation; training; words; Acoustic testing; Audio recording; Dogs; Extraterrestrial measurements; Horses; Indexing; Joining processes; Space shuttles; Surface fitting; Web pages;

fLanguage

English

Publisher

ieee

Conference_Titel

Multimedia and Expo, 2002. ICME '02. Proceedings. 2002 IEEE International Conference on

Print_ISBN

0-7803-7304-9

Type

conf

DOI

10.1109/ICME.2002.1035789

Filename

1035789