Text-dependent speaker identification based on input/output HMMs: An empirical study

Ke Chen; Dahong Xie; Huisheng Chi

Text-dependent speaker identification based on input/output HMMs: An empirical study

Ke Chen, Dahong Xie, Huisheng Chi

Research output: Chapter in Book/Conference proceeding › Conference contribution › peer-review

Abstract

In this paper, we explore the Input/Output HMM (IOHMM) architecture for a substantial problem, that of text-dependent speaker identification. For subnetworks modeled with generalized linear models, we extend the IRLS algorithm to the M-step of the corresponding EM algorithm. Experimental results show that the improved EM algorithm yields significantly faster training than the original one. In comparison with the multilayer perceptron, the dynamic programming technique and hidden Markov models, we empirically demonstrate that the IOHMM architecture is a promising way to text-dependent speaker identification. © 1996 Kluwer Academic Publishers.

Original language	English
Title of host publication	Neural Processing Letters\|Neural Process Letters
Pages	81-89
Number of pages	8
Volume	3
Publication status	Published - 1996
Event	Proceeding of World Congress on Neural Networks (WCNN-96) - San Diego, U.S.A. Duration: 1 Jan 1824 → …

Conference

Conference	Proceeding of World Congress on Neural Networks (WCNN-96)
City	San Diego, U.S.A.
Period	1/01/24 → …

Keywords

EM algorithm
Input/output HMM
Speaker identification
Temporal processing

Cite this

@inproceedings{c9f7dd7682b842ebb7b0161fb0e28eb9,

title = "Text-dependent speaker identification based on input/output HMMs: An empirical study",

abstract = "In this paper, we explore the Input/Output HMM (IOHMM) architecture for a substantial problem, that of text-dependent speaker identification. For subnetworks modeled with generalized linear models, we extend the IRLS algorithm to the M-step of the corresponding EM algorithm. Experimental results show that the improved EM algorithm yields significantly faster training than the original one. In comparison with the multilayer perceptron, the dynamic programming technique and hidden Markov models, we empirically demonstrate that the IOHMM architecture is a promising way to text-dependent speaker identification. {\textcopyright} 1996 Kluwer Academic Publishers.",

keywords = "EM algorithm, Input/output HMM, Speaker identification, Temporal processing",

author = "Ke Chen and Dahong Xie and Huisheng Chi",

year = "1996",

language = "English",

volume = "3",

pages = "81--89",

booktitle = "Neural Processing Letters|Neural Process Letters",

note = "Proceeding of World Congress on Neural Networks (WCNN-96) ; Conference date: 01-01-1824",

}

TY - GEN

T1 - Text-dependent speaker identification based on input/output HMMs: An empirical study

AU - Chen, Ke

AU - Xie, Dahong

AU - Chi, Huisheng

PY - 1996

Y1 - 1996

N2 - In this paper, we explore the Input/Output HMM (IOHMM) architecture for a substantial problem, that of text-dependent speaker identification. For subnetworks modeled with generalized linear models, we extend the IRLS algorithm to the M-step of the corresponding EM algorithm. Experimental results show that the improved EM algorithm yields significantly faster training than the original one. In comparison with the multilayer perceptron, the dynamic programming technique and hidden Markov models, we empirically demonstrate that the IOHMM architecture is a promising way to text-dependent speaker identification. © 1996 Kluwer Academic Publishers.

AB - In this paper, we explore the Input/Output HMM (IOHMM) architecture for a substantial problem, that of text-dependent speaker identification. For subnetworks modeled with generalized linear models, we extend the IRLS algorithm to the M-step of the corresponding EM algorithm. Experimental results show that the improved EM algorithm yields significantly faster training than the original one. In comparison with the multilayer perceptron, the dynamic programming technique and hidden Markov models, we empirically demonstrate that the IOHMM architecture is a promising way to text-dependent speaker identification. © 1996 Kluwer Academic Publishers.

KW - EM algorithm

KW - Input/output HMM

KW - Speaker identification

KW - Temporal processing

M3 - Conference contribution

VL - 3

SP - 81

EP - 89

BT - Neural Processing Letters|Neural Process Letters

T2 - Proceeding of World Congress on Neural Networks (WCNN-96)

Y2 - 1 January 1824

ER -

Text-dependent speaker identification based on input/output HMMs: An empirical study

Abstract

Conference

Keywords

Fingerprint

Cite this