Title :
Feature extraction for robust speech recognition
Author :
Dharanipragada, Satya
Author_Institution :
Dept. Human Language Technol., IBM Thomas J. Watson Res. Center, Yorktown Heights, NY, USA
Abstract :
We present new feature extraction and feature transformation techniques that significantly improve the robustness of continuous speech recognition. We report results on two tasks - an in-car embedded speech recognition task and a speaker-phone based air travel task. The techniques provide significant accuracy improvements in both tasks
Keywords :
feature extraction; spectral analysis; speech recognition; acoustic conditions; continuous speech recognition; feature extraction; feature transformation; handheld devices; in-car embedded speech recognition; minimum variance distortionless response; nonlinear feature transformation; robust speech recognition; speaker-phone based air travel task; spectrum estimation; speech input; Band pass filters; Cepstral analysis; Feature extraction; Handheld computers; Humans; Maximum likelihood linear regression; Natural languages; Noise robustness; Spectral analysis; Speech recognition;
Conference_Titel :
Circuits and Systems, 2002. ISCAS 2002. IEEE International Symposium on
Conference_Location :
Phoenix-Scottsdale, AZ
Print_ISBN :
0-7803-7448-7
DOI :
10.1109/ISCAS.2002.1011488
