Building a highly accurate mandarin speech recognizer

Mei Yuh Hwang; Gang Peng; Wen Wang; Arlo Faria; Aaron Heidel; Mari Ostendorf

Building a highly accurate mandarin speech recognizer

Mei Yuh Hwang, Gang Peng, Wen Wang, Arlo Faria, Aaron Heidel, Mari Ostendorf

Research output: Chapter in book / Conference proceeding › Conference article published in proceeding or book › Academic research › peer-review

20 Citations (Scopus)

Abstract

We describe a highly accurate large-vocabulary continuous Mandarin speech recognizer, a collaborative effort among four research organizations. Particularly, we build two acoustic models (AMs) with significant differences but similar accuracy for the purposes of cross adaptation and system combination. This paper elaborates on the main differences between the two systems, where one recognizer incorporates a discriminatively trained feature while the other utilizes a discriminative feature transformation. Additionally we present an improved acoustic segmentation algorithm and topicbased language model (LM) adaptation. Coupled with increased acoustic training data, we reduced the character error rate (CER) of the DARPA GALE 2006 evaluation set to 15.3% from 18.4%.

Original language	English
Title of host publication	2007 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2007, Proceedings
Pages	490-495
Number of pages	6
Publication status	Published - 1 Dec 2007
Externally published	Yes
Event	2007 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2007 - Kyoto, Japan Duration: 9 Dec 2007 → 13 Dec 2007

Conference

Conference	2007 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2007
Country/Territory	Japan
City	Kyoto
Period	9/12/07 → 13/12/07

Keywords

Acoustic segmentation
Character error rates
Discriminative features
LM adaptation
Mandarin
Multi-layer perceptrons
Out-of-vocabulary

ASJC Scopus subject areas

Computer Vision and Pattern Recognition
Software
Artificial Intelligence

Cite this

@inproceedings{cd9415adc3ce48cf95331284251fafe5,

title = "Building a highly accurate mandarin speech recognizer",

abstract = "We describe a highly accurate large-vocabulary continuous Mandarin speech recognizer, a collaborative effort among four research organizations. Particularly, we build two acoustic models (AMs) with significant differences but similar accuracy for the purposes of cross adaptation and system combination. This paper elaborates on the main differences between the two systems, where one recognizer incorporates a discriminatively trained feature while the other utilizes a discriminative feature transformation. Additionally we present an improved acoustic segmentation algorithm and topicbased language model (LM) adaptation. Coupled with increased acoustic training data, we reduced the character error rate (CER) of the DARPA GALE 2006 evaluation set to 15.3% from 18.4%.",

keywords = "Acoustic segmentation, Character error rates, Discriminative features, LM adaptation, Mandarin, Multi-layer perceptrons, Out-of-vocabulary",

author = "Hwang, {Mei Yuh} and Gang Peng and Wen Wang and Arlo Faria and Aaron Heidel and Mari Ostendorf",

year = "2007",

month = dec,

day = "1",

language = "English",

pages = "490--495",

booktitle = "2007 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2007, Proceedings",

note = "2007 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2007 ; Conference date: 09-12-2007 Through 13-12-2007",

}

Building a highly accurate mandarin speech recognizer. / Hwang, Mei Yuh; Peng, Gang; Wang, Wen et al.
2007 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2007, Proceedings. 2007. p. 490-495 4430161.

Research output: Chapter in book / Conference proceeding › Conference article published in proceeding or book › Academic research › peer-review

TY - GEN

T1 - Building a highly accurate mandarin speech recognizer

AU - Hwang, Mei Yuh

AU - Peng, Gang

AU - Wang, Wen

AU - Faria, Arlo

AU - Heidel, Aaron

AU - Ostendorf, Mari

PY - 2007/12/1

Y1 - 2007/12/1

N2 - We describe a highly accurate large-vocabulary continuous Mandarin speech recognizer, a collaborative effort among four research organizations. Particularly, we build two acoustic models (AMs) with significant differences but similar accuracy for the purposes of cross adaptation and system combination. This paper elaborates on the main differences between the two systems, where one recognizer incorporates a discriminatively trained feature while the other utilizes a discriminative feature transformation. Additionally we present an improved acoustic segmentation algorithm and topicbased language model (LM) adaptation. Coupled with increased acoustic training data, we reduced the character error rate (CER) of the DARPA GALE 2006 evaluation set to 15.3% from 18.4%.

AB - We describe a highly accurate large-vocabulary continuous Mandarin speech recognizer, a collaborative effort among four research organizations. Particularly, we build two acoustic models (AMs) with significant differences but similar accuracy for the purposes of cross adaptation and system combination. This paper elaborates on the main differences between the two systems, where one recognizer incorporates a discriminatively trained feature while the other utilizes a discriminative feature transformation. Additionally we present an improved acoustic segmentation algorithm and topicbased language model (LM) adaptation. Coupled with increased acoustic training data, we reduced the character error rate (CER) of the DARPA GALE 2006 evaluation set to 15.3% from 18.4%.

KW - Acoustic segmentation

KW - Character error rates

KW - Discriminative features

KW - LM adaptation

KW - Mandarin

KW - Multi-layer perceptrons

KW - Out-of-vocabulary

UR - http://www.scopus.com/inward/record.url?scp=44849087457&partnerID=8YFLogxK

M3 - Conference article published in proceeding or book

SP - 490

EP - 495

BT - 2007 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2007, Proceedings

T2 - 2007 IEEE Workshop on Automatic Speech Recognition and Understanding, ASRU 2007

Y2 - 9 December 2007 through 13 December 2007

ER -

Building a highly accurate mandarin speech recognizer

Abstract

Conference

Keywords

ASJC Scopus subject areas

Other files and links

Fingerprint

Cite this