東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Learning-based auditory encoding for...

Chiu, Yu-Hsiang Bosco.

Linked to FindBook

Google Book

Amazon

博客來

Learning-based auditory encoding for robust speech recognition.

Record Type:	Language materials, printed : Monograph/item
Title/Author:	Learning-based auditory encoding for robust speech recognition./
Author:	Chiu, Yu-Hsiang Bosco.
Description:	82 p.
Notes:	Source: Dissertation Abstracts International, Volume: 71-10, Section: B, page: 6219.
Contained By:	Dissertation Abstracts International71-10B.
Subject:	Engineering, Electronics and Electrical. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3421730
ISBN:	9781124217468

Learning-based auditory encoding for robust speech recognition.
Chiu, Yu-Hsiang Bosco.

Learning-based auditory encoding for robust speech recognition. - 82 p.

Source: Dissertation Abstracts International, Volume: 71-10, Section: B, page: 6219.

Thesis (Ph.D.)--Carnegie Mellon University, 2010.

While there has been a great deal of research in the area of automatic speech recognition (ASR) with substantial improvements in performance realized by current large vocabulary speech systems, the application of speech recognition to real environments remains limited because of serious degradation in accuracy. One of the most common causes for this loss of accuracy is a mismatch between training and testing environments. The goal of this thesis is to develop a set of new approaches to the signal processing used to extract features for speech recognition that are more robust to changes in the acoustical environment. We begin with an analysis of the relative effectiveness of the various stages of a popular physiologically-motivated model of feature extraction toward the improvement of recognition accuracy in the presence of additive noise. We then propose a new approach toward the extraction of speech features which is shown to be more robust to environmental distortion. Key parameters of the improved model are obtained using data-driven optimization rather by direct modeling of physiologically-measured data. In this work we focus our attention on (1) the nonlinear compressive function that relates the input signal level to the output level of neural activity in each frequency band, and (2) the modulation transfer function, which filters the filters that emerge from the output of the nonlinearity. Based on these analyses, we develop a set of algorithms that obtain the parameters that specify these modulation filters and rate-level nonlinearities. Finally, we discuss ways of reducing the computational complexity required to determine the optimal parameters for the feature extraction algorithms.

ISBN: 9781124217468Subjects--Topical Terms:

626636
Engineering, Electronics and Electrical.

Learning-based auditory encoding for robust speech recognition.
LDR:02535nam 2200265 4500 001 1395996
005 20110527105443.5
008 130515s2010 ||||||||||||||||| ||eng d
020 $a 9781124217468
035 $a (UMI)AAI3421730
035 $a AAI3421730
040 $a UMI $c UMI
100 1 $a Chiu, Yu-Hsiang Bosco. $3 1674745
245 1 0 $a Learning-based auditory encoding for robust speech recognition.
300 $a 82 p.
500 $a Source: Dissertation Abstracts International, Volume: 71-10, Section: B, page: 6219.
502 $a Thesis (Ph.D.)--Carnegie Mellon University, 2010.
520 $a While there has been a great deal of research in the area of automatic speech recognition (ASR) with substantial improvements in performance realized by current large vocabulary speech systems, the application of speech recognition to real environments remains limited because of serious degradation in accuracy. One of the most common causes for this loss of accuracy is a mismatch between training and testing environments. The goal of this thesis is to develop a set of new approaches to the signal processing used to extract features for speech recognition that are more robust to changes in the acoustical environment. We begin with an analysis of the relative effectiveness of the various stages of a popular physiologically-motivated model of feature extraction toward the improvement of recognition accuracy in the presence of additive noise. We then propose a new approach toward the extraction of speech features which is shown to be more robust to environmental distortion. Key parameters of the improved model are obtained using data-driven optimization rather by direct modeling of physiologically-measured data. In this work we focus our attention on (1) the nonlinear compressive function that relates the input signal level to the output level of neural activity in each frequency band, and (2) the modulation transfer function, which filters the filters that emerge from the output of the nonlinearity. Based on these analyses, we develop a set of algorithms that obtain the parameters that specify these modulation filters and rate-level nonlinearities. Finally, we discuss ways of reducing the computational complexity required to determine the optimal parameters for the feature extraction algorithms.
590 $a School code: 0041.
650 4 $a Engineering, Electronics and Electrical. $3 626636
650 4 $a Engineering, Robotics. $3 1018454
650 4 $a Computer Science. $3 626642
690 $a 0544
690 $a 0771
690 $a 0984
710 2 $a Carnegie Mellon University. $3 1018096
773 0 $t Dissertation Abstracts International $g 71-10B.
790 $a 0041
791 $a Ph.D.
792 $a 2010
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3421730