Title :
Spectral compensation for linear prediction of speech signals in coloured noise
Author_Institution :
Dept. of Electron. Eng., Nat. I-Lan Inst. of Agric. & Technol., Taiwan
fDate :
5/28/1998 12:00:00 AM
Abstract :
A method is presented to determine an appropriate portion that could be subtracted from the noise-contaminated autocorrelation functions before carrying out linear prediction analysis in coloured noise. This method guarantees the stability of the resulting linear prediction filter. The authors demonstrate the improvement using an objective measure based on a synthetic vowel
Keywords :
compensation; linear predictive coding; noise; speech processing; speech recognition; stability; coloured noise; linear prediction; linear prediction analysis; linear prediction filter; noise-contaminated autocorrelation functions; objective measure; spectral compensation; speech signals; stability; synthetic vowel;
Journal_Title :
Electronics Letters
DOI :
10.1049/el:19980814