Feature extraction based on minimum classification error/generalized probabilistic descent method

Authors:
Alain Biem;Shigeru Katagiri
Affiliations:
ATR Auditory and Visual Perception Research Laboratories;ATR Auditory and Visual Perception Research Laboratories
Venue:
ICASSP'93 Proceedings of the 1993 IEEE international conference on Acoustics, speech, and signal processing: speech processing - Volume II
Year:
1993

Citing 4
Cited 3

Prototype-based discriminative training for various speech units

ICASSP'92 Proceedings of the 1992 IEEE international conference on Acoustics, speech and signal processing - Volume 1
Segmental GPD training of HMM based speech recognizer

ICASSP'92 Proceedings of the 1992 IEEE international conference on Acoustics, speech and signal processing - Volume 1
Discriminative template training for dynamic programming speech recognition

ICASSP'92 Proceedings of the 1992 IEEE international conference on Acoustics, speech and signal processing - Volume 1
Application of a generalized probabilistic descent method to dynamic time warping-based speech recognition

ICASSP'92 Proceedings of the 1992 IEEE international conference on Acoustics, speech and signal processing - Volume 1

Optimized discriminative transformations for speech features based on minimum classification error

Pattern Recognition Letters
Discriminative feature extraction based on PCA Gaussian mixture models

DNCOCO'06 Proceedings of the 5th WSEAS international conference on Data networks, communications and computers
Discriminative feature extraction for speech recognition using continuous output codes

Pattern Recognition Letters

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper introduces a new approach to pattern recognition which comprehensively optimizes both a feature extraction process and a classification process. Here, assuming that the best features for recognition are the ones that yield the lowest classification error rate over unknown data, an overall recognizer, consisting of a feature extractor module and a classifier module, is trained using the Minimum Classification Error (MCE) with the Generalized Probabilistic Descent method (GPD). As an example of this MCE/GPD application, we specifically present experimental results of the proposed idea to a cepstrum liftering-based feature extraction apply to the vowel recognition. Comparison with conventional approaches shows how speech characteristics are extracted for recognition purposes.