线谱频率高效编码算法-研究.pdf

Abstract Abstract In the past twenty years, speech coding has become the one of most important key technologies in modern digital communication fields related to speech communications. Linear predictive coding model and its parameter line spectrum frequency (LSF) are widely used in various speech processing applications to represent the spectral envelope information of speech. The methods used for efficient line spectrum frequency quantization is vector quantization (VQ), of which the multi-stage VQ and split VQ commonly used. It is verified that with the same computational complexity, the multi-stage VQ is superior to the split VQ in performance, because in the multi-stage VQ scheme the simultaneous joint codebook design algorithm is used. In the thesis, a basic VQ scheme for LSF quantization based on LBG algorithm is designed firstly. Secondly, based on the LBG VQ design algorithm, a sequential multi-stage LSF VQ codebook design algorithm is implemented. Since the conjoint LSFs are quite similar in their values, so the inter-frame and intra-frame correlation of the LSFs can be used to predict the following and conjoint LSFs and remove the correlations between them. In this way, the performance of the LSF VQ can be improved. Thirdly, a multi-stage joint VQ codebook design algorithm and the related VQ coding algorithm are designed and implemented. In addition, two methods of multi-stage joint codebook with first-order autoregressive prediction and prediction of one-step interpolation are designed and implemented as well. Test and evaluation results have shown that, the performance in quality of the VQ codebook design algorithms and the related coding algorithms desi