Page 1 |
Save page Remove page | Previous | 1 of 185 | Next |
|
small (250x250 max)
medium (500x500 max)
Large (1000x1000 max)
Extra Large
large ( > 500x500)
Full Resolution
All (PDF)
|
This page
All
|
COMPRESSION ALGORITHMS FOR DISTRIBUTED CLASSIFICATION WITH
APPLICATIONS TO DISTRIBUTED SPEECH RECOGNITION
by
Naveen Srinivasamurthy
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ELECTRICAL ENGINEERING)
May 2007
Copyright 2007 Naveen Srinivasamurthy
Object Description
| Title | Compression algorithms for distributed classification with applications to distributed speech recognition |
| Author | Srinivasamurthy, Naveen |
| Author email | naveens@qualcomm.com |
| Degree | Doctor of Philosophy |
| Document type | Dissertation |
| Degree program | Electrical Engineering |
| School | Viterbi School of Engineering |
| Date defended/completed | 2003-09-10 |
| Date submitted | 2007 |
| Restricted until | Unrestricted |
| Date published | 2007-04-17 |
| Advisor (committee chair) | Ortega, Antonio |
| Advisor (committee member) |
Narayanan, Shrikanth S. Rozovsky, Boris |
| Abstract | With wide proliferation of mobile devices and the explosion of new multimedia applications, there is a need for client-server architectures to enable low complexity/memory clients to support complex multimedia applications. In these client-server systems compression is vital to minimize the transmission bandwidth requirements. Compression techniques have been designed to minimize perceptual distortion, however, recently there has been an emergence of applications in which an algorithm processes the compressed data. Here for best system performance the compression algorithm should be optimized to have the least effect on the algorithms estimation/classification capability. In this work novel compression techniques optimized forclassification are proposed.; The first application is remote speech recognition, where the speech recognizer recognizes spoken utterances using compressed data. A scalable encoder designed to maximize recognition performance is proposed, and shown to have superior rate-recognition performance than conventional speech encoders. A scalable recognition system capable of trading-off recognition performance for reduced complexity is proposed. These are useful in distributed speech recognition systems where several clients access a single server and efficient server design to reduce computational complexity and bandwidth requirements is required.; The second application is distributed classification, where the classifier operates on compressed data. A novel algorithm is proposed which is shown to significant reduce the misclassification penalty. The algorithm is extended to improve the performance of table-lookup encoders, where product vector quantizers are designed to approximate a higher dimension vector quantizer. Significant improvements in PSNR performance over conventional design is demonstrated with minimal increase in the encoding time.; Finally, a new distortion metric, mutual information loss (MIL), is proposed for designing quantizers in distributed classification applications. It is shown that the MI loss optimized quantizers are able to provide significant improvements in classification performance when compared to mean square error optimized quantizers. Empirical quantizer design and rate allocation algorithms are provided to optimize quantizers for minimizing MI loss. Additionally, it is shown that the MI loss metric can be used to design quantizers operating on low dimension vectors. This is a vital requirement in classification systems employing high dimension classifiers as it enables design of optimal and practical minimum MI loss quantizers implementable on low complexity/memory clients. |
| Keyword | distributed classification; distributed speech recognition, Joint compression and classification; DPCM; quantization; scalable encoding; scalable DPCM; product vector quantization; mutual information loss; information bottleneck |
| Language | English |
| Part of collection | University of Southern California dissertations and theses |
| Publisher (of the original version) | University of Southern California |
| Place of publication (of the original version) | Los Angeles, California |
| Publisher (of the digital version) | University of Southern California. Libraries |
| Type | texts |
| Legacy record ID | usctheses-m401 |
| Contributing entity | University of Southern California |
| Rights | Srinivasamurthy, Naveen |
| Repository name | Libraries, University of Southern California |
| Repository address | Los Angeles, California |
| Repository email | cisadmin@lib.usc.edu |
| Filename | etd-Srinivasamurthy-20070417 |
| Archival file | uscthesesreloadpub_Volume14/etd-Srinivasamurthy-20070417.pdf |
Description
| Title | Page 1 |
| Contributing entity | University of Southern California |
| Repository email | cisadmin@lib.usc.edu |
| Full text | COMPRESSION ALGORITHMS FOR DISTRIBUTED CLASSIFICATION WITH APPLICATIONS TO DISTRIBUTED SPEECH RECOGNITION by Naveen Srinivasamurthy A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (ELECTRICAL ENGINEERING) May 2007 Copyright 2007 Naveen Srinivasamurthy |
