HMM adaptation techniques in training framework

Sam KWONG; Qianhua HE; Y. K. CHAN

doi:10.1109/TENCON.2001.949612

HMM adaptation techniques in training framework

Sam KWONG, Qianhua HE, Y. K. CHAN

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Research › peer-review

Abstract

This paper presented an adaptation approach based on Baum-Welch algorithm method. This method applies the same framework as they are used for training speech recognizers with abundant training data. The Baum-Welch adaptation method adapted to all the parameters of the hidden Markov models (HMM) with adaptation data. If large amount of adaptation data is available, these methods could gradually approximate the speaker-dependent ones. The approach is evaluated through the phoneme recognition task on the TIMIT corpus. On the speaker adaptation experiments, up to 91.48% recognition rate is achieved.

Original language	English
Title of host publication	Proceedings of IEEE Region 10 International Conference on Electrical and Electronic Technology
Editors	Dapeng TIEN, Yung C. LIANG
Publisher	IEEE
Pages	350-354
Number of pages	5
DOIs	https://doi.org/10.1109/TENCON.2001.949612
Publication status	Published - 2001
Externally published	Yes
Event	IEEE Region 10 International Conference on Electrical and Electronic Technology - , Singapore Duration: 19 Aug 2001 → 22 Aug 2001

Conference

Conference	IEEE Region 10 International Conference on Electrical and Electronic Technology
Country/Territory	Singapore
Period	19/08/01 → 22/08/01

Keywords

And Maximum Likelihood
Hidden Markov Model
Maximum Model Distance
Speaker Adaptation

Access to Document

10.1109/TENCON.2001.949612

Cite this

@inproceedings{64dc202feda84a4f86836f7e49882cef,

title = "HMM adaptation techniques in training framework",

abstract = "This paper presented an adaptation approach based on Baum-Welch algorithm method. This method applies the same framework as they are used for training speech recognizers with abundant training data. The Baum-Welch adaptation method adapted to all the parameters of the hidden Markov models (HMM) with adaptation data. If large amount of adaptation data is available, these methods could gradually approximate the speaker-dependent ones. The approach is evaluated through the phoneme recognition task on the TIMIT corpus. On the speaker adaptation experiments, up to 91.48% recognition rate is achieved.",

keywords = "And Maximum Likelihood, Hidden Markov Model, Maximum Model Distance, Speaker Adaptation",

author = "Sam KWONG and Qianhua HE and CHAN, {Y. K.}",

year = "2001",

doi = "10.1109/TENCON.2001.949612",

language = "English",

pages = "350--354",

editor = "Dapeng TIEN and LIANG, {Yung C.}",

booktitle = "Proceedings of IEEE Region 10 International Conference on Electrical and Electronic Technology",

publisher = "IEEE",

note = "IEEE Region 10 International Conference on Electrical and Electronic Technology ; Conference date: 19-08-2001 Through 22-08-2001",

}

TY - GEN

T1 - HMM adaptation techniques in training framework

AU - KWONG, Sam

AU - HE, Qianhua

AU - CHAN, Y. K.

PY - 2001

Y1 - 2001

N2 - This paper presented an adaptation approach based on Baum-Welch algorithm method. This method applies the same framework as they are used for training speech recognizers with abundant training data. The Baum-Welch adaptation method adapted to all the parameters of the hidden Markov models (HMM) with adaptation data. If large amount of adaptation data is available, these methods could gradually approximate the speaker-dependent ones. The approach is evaluated through the phoneme recognition task on the TIMIT corpus. On the speaker adaptation experiments, up to 91.48% recognition rate is achieved.

AB - This paper presented an adaptation approach based on Baum-Welch algorithm method. This method applies the same framework as they are used for training speech recognizers with abundant training data. The Baum-Welch adaptation method adapted to all the parameters of the hidden Markov models (HMM) with adaptation data. If large amount of adaptation data is available, these methods could gradually approximate the speaker-dependent ones. The approach is evaluated through the phoneme recognition task on the TIMIT corpus. On the speaker adaptation experiments, up to 91.48% recognition rate is achieved.

KW - And Maximum Likelihood

KW - Hidden Markov Model

KW - Maximum Model Distance

KW - Speaker Adaptation

UR - http://www.scopus.com/inward/record.url?scp=0035786199&partnerID=8YFLogxK

U2 - 10.1109/TENCON.2001.949612

DO - 10.1109/TENCON.2001.949612

M3 - Conference paper (refereed)

SP - 350

EP - 354

BT - Proceedings of IEEE Region 10 International Conference on Electrical and Electronic Technology

A2 - TIEN, Dapeng

A2 - LIANG, Yung C.

PB - IEEE

T2 - IEEE Region 10 International Conference on Electrical and Electronic Technology

Y2 - 19 August 2001 through 22 August 2001

ER -

HMM adaptation techniques in training framework

Abstract

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this