Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

A Perception and PDE Based Nonlinear Transformation for Processing Spoken Words

Yingyong Qi (1), Jack Xin (2)

(1) Dept. of Electrical Computer Engineering, University of Arizona, Tucson, AZ, USA
(2)Department of Mathematics and TICAM, University of Texas at Austin, USA

A perception and PDE (partial di erential equa- tion) based nonlinear transformation is proposed to process spoken words in noisy environment. The transformation was designed to reduce noise through time adaptation and spectral enhancement by evolving a focusing quadratic fourth order nonlinear PDE (the Cahn-Hillard equation). Constant, low SNR signals were adaptively decreased to reduce noise. Once the noise levels were reduced, the evolution of focusing PDE was used to enhance the spectral peaks, and further reduce noise interference. Numerical results on noisy spoken words indicated that the transformed spectral pattern of the spoken words was insensitive to nosie. The spectral distances between noisy words and original words were significantly reduced after the transformation.


Full Paper

Bibliographic reference.  Qi, Yingyong / Xin, Jack (2000): "A perception and PDE based nonlinear transformation for processing spoken words", In ICSLP-2000, vol.1, 445-448.