Title :
Integrated MMSE-FW-LA scheme for robust ASR at low SNRs
Author :
Xu, Tao ; Cao, Zhigang
Author_Institution :
Dept. of Electron. Eng., Tsinghua Univ., Beijing, China
fDate :
10/24/2002 12:00:00 AM
Abstract :
A novel feature weight (FW) algorithm and an integrated scheme called MMSE-FW-LA are proposed. The scheme combines the minimum mean square error (MMSE)-based enhancement, the FW algorithm and the log-add (LA) model compensation. Experimental evaluations show that this scheme can improve the recognition performance significantly, especially at low SNRs.
Keywords :
cepstral analysis; least mean squares methods; speech enhancement; speech recognition; MMSE-FW-LA scheme; automatic speech recognition; feature weight algorithm; integrated scheme; log-add model compensation; low SNR; minimum mean square error-based enhancement; robust ASR;
Journal_Title :
Electronics Letters
DOI :
10.1049/el:20020940
