Feature vs. model based vocal tract length normalization for a speech recognition-based interactive toy

Chun Keung Chau; Chak Shun Lai; Bertram Emil Shi

doi:10.1007/3-540-45336-9_17

Feature vs. model based vocal tract length normalization for a speech recognition-based interactive toy

Chun Keung Chau, Chak Shun Lai, Bertram Emil Shi

HKUST Fok Ying Tung Graduate School

Research output: Chapter in Book/Conference Proceeding/Report › Conference Paper published in a book › peer-review

1 Citation (Scopus)

Abstract

We describe an architecture for speech recognition based interactive toys and discuss the strategies we have adopted to deal with the requirements for the speech recognizer imposed by this application. In particular, we focus on the fact that speech recognizers used in interactive toys must deal with users whose age ranges from children to adults. The large variations in vocal tract length between children and adults can significantly degrade the performance of speech recognizers. We compare two approaches to vocal tract length normalization: feature-based VTLN and model-based VTLN. We describe why intuitively, one might expect that due to the coarser frequency information used by the model-based approach, that feature-based VTLN would outperform modelbased VTLN. However, our results indicate that there is very little difference in performance between the two schemes.

Original language	English
Title of host publication	Active Media Technology - 6th International Computer Science Conference, AMT 2001, Proceedings
Editors	Jiming Liu, Pong C. Yuen, Chun-hung Li, Joseph Ng, Toru Ishida
Publisher	Springer Verlag
Pages	134-143
Number of pages	10
ISBN (Electronic)	9783540430353
DOIs	https://doi.org/10.1007/3-540-45336-9_17
Publication status	Published - 2001
Event	6th International Computer Science Conference on Active Media Technology, AMT 2001 - Hong Kong, China Duration: 18 Dec 2001 → 20 Dec 2001

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	2252
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	6th International Computer Science Conference on Active Media Technology, AMT 2001
Country/Territory	China
City	Hong Kong
Period	18/12/01 → 20/12/01

Bibliographical note

Publisher Copyright:
© Springer-Verlag Berlin Heidelberg 2001.

Access to Document

10.1007/3-540-45336-9_17

Cite this

Chau, C. K., Lai, C. S., & Shi, B. E. (2001). Feature vs. model based vocal tract length normalization for a speech recognition-based interactive toy. In J. Liu, P. C. Yuen, C. Li, J. Ng, & T. Ishida (Eds.), Active Media Technology - 6th International Computer Science Conference, AMT 2001, Proceedings (pp. 134-143). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 2252). Springer Verlag. https://doi.org/10.1007/3-540-45336-9_17

Chau, Chun Keung ; Lai, Chak Shun ; Shi, Bertram Emil. / Feature vs. model based vocal tract length normalization for a speech recognition-based interactive toy. Active Media Technology - 6th International Computer Science Conference, AMT 2001, Proceedings. editor / Jiming Liu ; Pong C. Yuen ; Chun-hung Li ; Joseph Ng ; Toru Ishida. Springer Verlag, 2001. pp. 134-143 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{6f5fc25034a04fadb3a19178561a57b9,

title = "Feature vs. model based vocal tract length normalization for a speech recognition-based interactive toy",

abstract = "We describe an architecture for speech recognition based interactive toys and discuss the strategies we have adopted to deal with the requirements for the speech recognizer imposed by this application. In particular, we focus on the fact that speech recognizers used in interactive toys must deal with users whose age ranges from children to adults. The large variations in vocal tract length between children and adults can significantly degrade the performance of speech recognizers. We compare two approaches to vocal tract length normalization: feature-based VTLN and model-based VTLN. We describe why intuitively, one might expect that due to the coarser frequency information used by the model-based approach, that feature-based VTLN would outperform modelbased VTLN. However, our results indicate that there is very little difference in performance between the two schemes.",

author = "Chau, \{Chun Keung\} and Lai, \{Chak Shun\} and Shi, \{Bertram Emil\}",

note = "Publisher Copyright: {\textcopyright} Springer-Verlag Berlin Heidelberg 2001.; 6th International Computer Science Conference on Active Media Technology, AMT 2001 ; Conference date: 18-12-2001 Through 20-12-2001",

year = "2001",

doi = "10.1007/3-540-45336-9\_17",

language = "English",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "134--143",

editor = "Jiming Liu and Yuen, \{Pong C.\} and Chun-hung Li and Joseph Ng and Toru Ishida",

booktitle = "Active Media Technology - 6th International Computer Science Conference, AMT 2001, Proceedings",

}

Chau, CK, Lai, CS & Shi, BE 2001, Feature vs. model based vocal tract length normalization for a speech recognition-based interactive toy. in J Liu, PC Yuen, C Li, J Ng & T Ishida (eds), Active Media Technology - 6th International Computer Science Conference, AMT 2001, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 2252, Springer Verlag, pp. 134-143, 6th International Computer Science Conference on Active Media Technology, AMT 2001, Hong Kong, China, 18/12/01. https://doi.org/10.1007/3-540-45336-9_17

Feature vs. model based vocal tract length normalization for a speech recognition-based interactive toy. / Chau, Chun Keung; Lai, Chak Shun; Shi, Bertram Emil.
Active Media Technology - 6th International Computer Science Conference, AMT 2001, Proceedings. ed. / Jiming Liu; Pong C. Yuen; Chun-hung Li; Joseph Ng; Toru Ishida. Springer Verlag, 2001. p. 134-143 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 2252).

Research output: Chapter in Book/Conference Proceeding/Report › Conference Paper published in a book › peer-review

TY - GEN

T1 - Feature vs. model based vocal tract length normalization for a speech recognition-based interactive toy

AU - Chau, Chun Keung

AU - Lai, Chak Shun

AU - Shi, Bertram Emil

N1 - Publisher Copyright: © Springer-Verlag Berlin Heidelberg 2001.

PY - 2001

Y1 - 2001

N2 - We describe an architecture for speech recognition based interactive toys and discuss the strategies we have adopted to deal with the requirements for the speech recognizer imposed by this application. In particular, we focus on the fact that speech recognizers used in interactive toys must deal with users whose age ranges from children to adults. The large variations in vocal tract length between children and adults can significantly degrade the performance of speech recognizers. We compare two approaches to vocal tract length normalization: feature-based VTLN and model-based VTLN. We describe why intuitively, one might expect that due to the coarser frequency information used by the model-based approach, that feature-based VTLN would outperform modelbased VTLN. However, our results indicate that there is very little difference in performance between the two schemes.

AB - We describe an architecture for speech recognition based interactive toys and discuss the strategies we have adopted to deal with the requirements for the speech recognizer imposed by this application. In particular, we focus on the fact that speech recognizers used in interactive toys must deal with users whose age ranges from children to adults. The large variations in vocal tract length between children and adults can significantly degrade the performance of speech recognizers. We compare two approaches to vocal tract length normalization: feature-based VTLN and model-based VTLN. We describe why intuitively, one might expect that due to the coarser frequency information used by the model-based approach, that feature-based VTLN would outperform modelbased VTLN. However, our results indicate that there is very little difference in performance between the two schemes.

UR - https://openalex.org/W1521081523

UR - https://www.scopus.com/pages/publications/84931837769

U2 - 10.1007/3-540-45336-9_17

DO - 10.1007/3-540-45336-9_17

M3 - Conference Paper published in a book

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 134

EP - 143

BT - Active Media Technology - 6th International Computer Science Conference, AMT 2001, Proceedings

A2 - Liu, Jiming

A2 - Yuen, Pong C.

A2 - Li, Chun-hung

A2 - Ng, Joseph

A2 - Ishida, Toru

PB - Springer Verlag

T2 - 6th International Computer Science Conference on Active Media Technology, AMT 2001

Y2 - 18 December 2001 through 20 December 2001

ER -

Chau CK, Lai CS, Shi BE. Feature vs. model based vocal tract length normalization for a speech recognition-based interactive toy. In Liu J, Yuen PC, Li C, Ng J, Ishida T, editors, Active Media Technology - 6th International Computer Science Conference, AMT 2001, Proceedings. Springer Verlag. 2001. p. 134-143. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/3-540-45336-9_17

Feature vs. model based vocal tract length normalization for a speech recognition-based interactive toy

Abstract

Publication series

Conference

Bibliographical note

Access to Document

Other files and links

Fingerprint

Cite this