Share Email Print
cover

Proceedings Paper

Matching pursuit decomposition of speech signals for compact representation
Author(s): Ye Shen; Hongmei Ai; C.-C. Jay Kuo
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

Matching Pursuit (MP) expands a signal over an overcomplete dictionary of normalized atoms in an iterative fashion. A careful selection of dictionary components is critical in the design of the MP algorithm for compact signal representation and manipulation. In this research, the use of MP as an alternative waveform-coding scheme for speech signals is investigated. The improvement of MP over conventional transform coding schemes is due to the use of overcomplete basis functions. Furthermore, the performance of MP representation can be enhanced via a compact MP dictionary obtained from training. Inspired by the popular Vector Quantization (VQ) algorithm, a dictionary-training algorithm is proposed in this paper to find the optimal dictionary for MP in speech coding. The MP decomposition with a trained dictionary is shown to improve the compactness of speech representation over the traditional MP decomposition with a generic Gabor dictionary. A better SNR performance is achieved with a dictionary of a limited size, which has a good potential for future appliations.

Paper Details

Date Published: 10 December 2002
PDF: 10 pages
Proc. SPIE 4861, Multimedia Systems and Applications V, (10 December 2002); doi: 10.1117/12.456500
Show Author Affiliations
Ye Shen, Univ. of Southern California (United States)
Hongmei Ai, Univ. of Southern California (United States)
C.-C. Jay Kuo, Univ. of Southern California (United States)


Published in SPIE Proceedings Vol. 4861:
Multimedia Systems and Applications V
Andrew G. Tescher; Bhaskaran Vasudev; V. Michael Bove; Ajay Divakaran, Editor(s)

© SPIE. Terms of Use
Back to Top