An efficient adaptive LSP method for speech encoding

Author

Ching, P.C. ; Ho, K.C.

Author_Institution

Dept. of Electron. Eng., Chinese Univ. of Hong Kong, Shatin, Hong Kong

fYear

1990

fDate

24-27 Sep 1990

Firstpage

324

Abstract

The authors describe an adaptive split-path filter structure for calculating line spectral pair (LSP) coefficients by using the least-mean-squares (LMS) algorithm. The two split filters are constructed as a cascade of second-order sections and the filter parameters are equivalent to the LSP coefficients. Adaptation is based on minimizing the output errors of the two filters independently, which is computationally simpler and has a much faster convergence rate. A subband predictive coder that divides speech into three subbands, each encoded with fourth-order LSP coefficients representing each of the three predictive filters, has been simulated on an IBM PC. Preliminary results indicate that output speech produced at 12 to 16 kb/s is highly intelligible and has a quality comparable to the 16 kb/s continuously variable slope delta (CVSD) method

Keywords

encoding; filtering and prediction theory; speech analysis and processing; 12 to 16 kbit/s; LMS algorithm; LSP coefficients; adaptive split-path filter structure; convergence rate; least-mean-squares; line spectral pair; output errors; predictive filters; speech encoding; speech intelligibility; speech quality; subband predictive coder; Adaptive filters; Bit rate; Convergence; Encoding; Signal synthesis; Speech analysis; Speech coding; Speech enhancement; Speech synthesis; Transversal filters;

fLanguage

English

Publisher

ieee

Conference_Titel

Computer and Communication Systems, 1990. IEEE TENCON'90., 1990 IEEE Region 10 Conference on

Print_ISBN

0-87942-556-3

Type

conf

DOI

10.1109/TENCON.1990.152625

Filename

152625