Efficient embedded speech recognition for very large vocabulary Mandarin car-navigation systems

Yanmin Qian; Jia Liu; Michael T. Johnson

doi:10.1109/TCE.2009.5278018

Efficient embedded speech recognition for very large vocabulary Mandarin car-navigation systems

Yanmin Qian
, Jia Liu
, Michael T. Johnson

Research output: Contribution to journal › Article › peer-review

11 Scopus citations

Abstract

Automatic speech recognition (ASR) for a very large vocabulary of isolated words is a difficult task on a resource-limited embedded device. This paper presents a novel fast decoding algorithm for a mandarin speech recognition system which can simultaneously process hundreds of thousands of items and maintain high recognition accuracy. The proposed algorithm constructs a semi-tree search network based on mandarin pronunciation rules, to avoid duplicate syllable matching and save redundant memory. Based on a two-stage fixed-width beam-search baseline system, the algorithm employs a variable beam-width pruning strategy and a frame-synchronous word-level pruning strategy to significantly reduce recognition time. This algorithm is aimed at an in-car navigation system in China and simulated on a standard PC workstation. The experimental results show that the proposed method reduces recognition time by nearly 6-fold and memory size nearly 2- fold compared to the baseline system, and causes less than 1% accuracy degradation for a 200,000 word recognition task.

Original language	English
Pages (from-to)	1496-1500
Number of pages	5
Journal	IEEE Transactions on Consumer Electronics
Volume	55
Issue number	3
DOIs	https://doi.org/10.1109/TCE.2009.5278018
State	Published - 2009

Bibliographical note

Funding Information:
1This work was supported by the National High Technology Research and Development Program of China (NO. 2006AA010101 and NO. 2007AA04Z223) and National Natural Science Foundation of China and Microsoft Research Asia (NO. 60776800). Yanmin Qian and Jia Liu are with the Department of Electronic Engineering, Tsinghua University, Beijing 100084, China (e-mail: [email protected]). Michael T. Johnson is with the Department of Electrical Engineering, Marquette University, Milwaukee, Wisconsin 53201, USA. He is now a visiting professor with the Department of Electronic Engineering, Tsinghua University, Beijing 100084, China (e-mail: [email protected]). Contributed Paper Manuscript received June 11, 2009 0098 3063/09/$20.00 © 2009 IEEE

Funding

1This work was supported by the National High Technology Research and Development Program of China (NO. 2006AA010101 and NO. 2007AA04Z223) and National Natural Science Foundation of China and Microsoft Research Asia (NO. 60776800). Yanmin Qian and Jia Liu are with the Department of Electronic Engineering, Tsinghua University, Beijing 100084, China (e-mail: [email protected]). Michael T. Johnson is with the Department of Electrical Engineering, Marquette University, Milwaukee, Wisconsin 53201, USA. He is now a visiting professor with the Department of Electronic Engineering, Tsinghua University, Beijing 100084, China (e-mail: [email protected]). Contributed Paper Manuscript received June 11, 2009 0098 3063/09/$20.00 © 2009 IEEE

Funders	Funder number
Microsoft Research Asia	60776800
National High Technology Research and Development Program of China	2006AA010101, 2007AA04Z223
National Natural Science Foundation of China (NSFC)

Keywords

Beam-search
Search network
Speech recognition
Word-level pruning

ASJC Scopus subject areas

Media Technology
Electrical and Electronic Engineering

Access to Document

10.1109/TCE.2009.5278018

Cite this

@article{77ac1098bb1e4e33b817fa62a4c3712f,

title = "Efficient embedded speech recognition for very large vocabulary Mandarin car-navigation systems",

abstract = "Automatic speech recognition (ASR) for a very large vocabulary of isolated words is a difficult task on a resource-limited embedded device. This paper presents a novel fast decoding algorithm for a mandarin speech recognition system which can simultaneously process hundreds of thousands of items and maintain high recognition accuracy. The proposed algorithm constructs a semi-tree search network based on mandarin pronunciation rules, to avoid duplicate syllable matching and save redundant memory. Based on a two-stage fixed-width beam-search baseline system, the algorithm employs a variable beam-width pruning strategy and a frame-synchronous word-level pruning strategy to significantly reduce recognition time. This algorithm is aimed at an in-car navigation system in China and simulated on a standard PC workstation. The experimental results show that the proposed method reduces recognition time by nearly 6-fold and memory size nearly 2- fold compared to the baseline system, and causes less than 1\% accuracy degradation for a 200,000 word recognition task.",

keywords = "Beam-search, Search network, Speech recognition, Word-level pruning",

author = "Yanmin Qian and Jia Liu and Johnson, \{Michael T.\}",

note = "Funding Information: 1This work was supported by the National High Technology Research and Development Program of China (NO. 2006AA010101 and NO. 2007AA04Z223) and National Natural Science Foundation of China and Microsoft Research Asia (NO. 60776800). Yanmin Qian and Jia Liu are with the Department of Electronic Engineering, Tsinghua University, Beijing 100084, China (e-mail: [email protected]). Michael T. Johnson is with the Department of Electrical Engineering, Marquette University, Milwaukee, Wisconsin 53201, USA. He is now a visiting professor with the Department of Electronic Engineering, Tsinghua University, Beijing 100084, China (e-mail: [email protected]). Contributed Paper Manuscript received June 11, 2009 0098 3063/09/\$20.00 {\textcopyright} 2009 IEEE",

year = "2009",

doi = "10.1109/TCE.2009.5278018",

language = "English",

volume = "55",

pages = "1496--1500",

number = "3",

}

TY - JOUR

T1 - Efficient embedded speech recognition for very large vocabulary Mandarin car-navigation systems

AU - Qian, Yanmin

AU - Liu, Jia

AU - Johnson, Michael T.

N1 - Funding Information: 1This work was supported by the National High Technology Research and Development Program of China (NO. 2006AA010101 and NO. 2007AA04Z223) and National Natural Science Foundation of China and Microsoft Research Asia (NO. 60776800). Yanmin Qian and Jia Liu are with the Department of Electronic Engineering, Tsinghua University, Beijing 100084, China (e-mail: [email protected]). Michael T. Johnson is with the Department of Electrical Engineering, Marquette University, Milwaukee, Wisconsin 53201, USA. He is now a visiting professor with the Department of Electronic Engineering, Tsinghua University, Beijing 100084, China (e-mail: [email protected]). Contributed Paper Manuscript received June 11, 2009 0098 3063/09/$20.00 © 2009 IEEE

PY - 2009

Y1 - 2009

N2 - Automatic speech recognition (ASR) for a very large vocabulary of isolated words is a difficult task on a resource-limited embedded device. This paper presents a novel fast decoding algorithm for a mandarin speech recognition system which can simultaneously process hundreds of thousands of items and maintain high recognition accuracy. The proposed algorithm constructs a semi-tree search network based on mandarin pronunciation rules, to avoid duplicate syllable matching and save redundant memory. Based on a two-stage fixed-width beam-search baseline system, the algorithm employs a variable beam-width pruning strategy and a frame-synchronous word-level pruning strategy to significantly reduce recognition time. This algorithm is aimed at an in-car navigation system in China and simulated on a standard PC workstation. The experimental results show that the proposed method reduces recognition time by nearly 6-fold and memory size nearly 2- fold compared to the baseline system, and causes less than 1% accuracy degradation for a 200,000 word recognition task.

AB - Automatic speech recognition (ASR) for a very large vocabulary of isolated words is a difficult task on a resource-limited embedded device. This paper presents a novel fast decoding algorithm for a mandarin speech recognition system which can simultaneously process hundreds of thousands of items and maintain high recognition accuracy. The proposed algorithm constructs a semi-tree search network based on mandarin pronunciation rules, to avoid duplicate syllable matching and save redundant memory. Based on a two-stage fixed-width beam-search baseline system, the algorithm employs a variable beam-width pruning strategy and a frame-synchronous word-level pruning strategy to significantly reduce recognition time. This algorithm is aimed at an in-car navigation system in China and simulated on a standard PC workstation. The experimental results show that the proposed method reduces recognition time by nearly 6-fold and memory size nearly 2- fold compared to the baseline system, and causes less than 1% accuracy degradation for a 200,000 word recognition task.

KW - Beam-search

KW - Search network

KW - Speech recognition

KW - Word-level pruning

UR - https://www.scopus.com/pages/publications/70350300962

UR - https://www.scopus.com/pages/publications/70350300962#tab=citedBy

U2 - 10.1109/TCE.2009.5278018

DO - 10.1109/TCE.2009.5278018

M3 - Article

AN - SCOPUS:70350300962

SN - 0098-3063

VL - 55

SP - 1496

EP - 1500

JO - IEEE Transactions on Consumer Electronics

JF - IEEE Transactions on Consumer Electronics

IS - 3

ER -

Efficient embedded speech recognition for very large vocabulary Mandarin car-navigation systems

Abstract

Bibliographical note

Funding

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this