CONSTRAINED TRANSITION VECTOR QUANTIZATION METHOD FOR NOISY CHANNEL SPEECH CODING

A novel coding method is considered here that combines source coding o f speech and channel coding. A vector quantization method using a cons trained transition codebook is proposed. Encoding with the proposed me thod gives codes an error-correction ability without addition of expli cit redundant bits by error-correcting coding. To compare the proposed and the ordinary source-channel tandem coding methods, we carried out computer simulations, transmitting LPC cepstrum vectors of speech. In the presence of channel errors, cepstrum distortion of the decoded ce pstrum vector by the proposed method is significantly decreased compar ed with one by the ordinary vector quantization. Moreover, from compar isons with tandem methods, it is shown that the proposed method is mor e robust than the ordinary tandem methods. Moreover, its error-correct ion ability is almost equal to rate 1/2 convolutional coding.