Classified Conditional Entropy Coding of LSP Parameters

In this paper, a new LSP speech parameter compression scheme is proposed which uses conditional probability information through classification. For efficient compression of speech LSP parameter vectors it is essential that higher order correlations are exploited. The use of conditional probability information has been hindered by high complexity of the information. For example, a LSP vector has 34 bit representation at 4.8 K bps CELP coding (FS1016 standard). It is impractical to use the first order probability information directly since 234 M 1.7 x 10” number of probability tables would be required and training of such information would be practically impossible. In order to reduce the complexity, we reduce the input alphabet size by classifying the LSP vectors according to their phonetic relevance. In other words, speech LSP parameters are classified into groups representing loosely defined various phonemes. The number of phoneme groups used was 32 considering ,the ambiguity of similar phonemes and background noises. Then conditional probability tables are constructed for each class by training. In order to further reduce the complexity, split-VQ has been employed [l]. The classification is achieved through vector quantization with a mean squared distortion measure in the LSP domain. First, a vector quantization of LSP parameters are done using 2-split or 3-split VQ methods. Next, the first order conditional entropy coding of sub-vectors is performed using the simplified conditional probability information. Since the number of conditioning states is very small, the required training data size is significantly reduced. Yet the performance penalty was very small. In our simulation, using frame size of 30 ms we were able to achieve less tban 1 dB spectral distortion at 21 bits/frame for LSP parameters which were originally coded with 34 bits/frame. It has been shown that the proposed approach outperforms a recent complexity reduction scheme called relative index coding (RIC).[Z] [l] K. K. Paliwal and B. S. Atal,“Efficient Vector Q uantization of LPC Parameters at 24 Bits/Frame”,IEEE Trans. on Speech and Audio Processing, Vol. 1, No. 1, pp. 3-14, January 1993. [2] S. Bruhn,“Efficient Interblock Noiseless Coding of Speech LPC parameters”, in Proc. ICASSP, 1-501-I-504, 1994. ‘Corresponding Author 1068-0314/95$4.00