TR-I-0223 :1991

Masahide SUGIYAMA, Shigeki SAGAYAMA

A Study on Noise Robust VQ Codebook Generation

Abstract:This report describes a preliminary study on noise robust VQ codebook generation. VQ technique is widely used in various speech fields. When VQ codes are encoded and transmitted with bit errors, the decoded VQ codes produce transmission error. This error produces the distortion of speech. The aim of this study is to generate robust VQ codebook. In this report, the problem is formulated and the several preliminary experiments are described. The results show that the generated robust VQ codebook performs 2 or 3dB reduction. This results is very encouraging. In the last part, Neural Networks applied to the optimal solution search algorithm will be explained.