Improving the training and testing speed and the ability of generalization in learning vector quantization-DVQ

Author

Poirier, Franck

Author_Institution

Telecom Paris, France

fYear

1991

fDate

14-17 Apr 1991

Firstpage

649

Abstract

Learning vector quantization (LVQ) is a nearest neighbor classifier very close to the self-organizing feature map classifier. A novel method called dynamic vector quantization (DVQ) is proposed for improving the ability to generalize and the learning and testing speed. DVQ is evaluated on speech data and synthetic data. DVQ always gives best results with fewer reference vectors than LVQ2. On speech experiments, DVQ shows an improvement of about 5% in the recognition rate, and the learning speed is three times faster

Keywords

data compression; learning systems; neural nets; speech recognition; DVQ; LVQ; dynamic vector quantization; learning speed; learning vector quantization; nearest neighbor classifier; neural networks; recognition rate; reference vectors; speech data; speech experiments; speech recognition; synthetic data; testing speed; training speed; Acoustic testing; Databases; Decoding; Gaussian distribution; Hidden Markov models; Loudspeakers; Neural networks; Speech analysis; Speech recognition; Vector quantization;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference on

Conference_Location

Toronto, Ont.

ISSN

1520-6149

Print_ISBN

0-7803-0003-3

Type

conf

DOI

10.1109/ICASSP.1991.150423

Filename

150423