Gmm-based Low Bit Rate Speech Coding,

A novel low bit-rate speech coder based on Gaussian Mixture Model (GMM), which is used to parameterize the short-time speech spectrum envelope, is researched in this paper. Since the segmented speech can be represented by very few parameters of GMM, the bit-rate of the coder is very low.The spectrum envelope carries very important information of the speech. Different methods as LPC, LPC cepstrum and SEEVOC for obtaining the spectrum envelope are analyzed. By some comparisons, the method SEEVOC is utilized. Then the spectrum envelope can be represented by the means, covariances and mixture weights of GMM.The pitch affects the quality of synthesized speech. A modified pitch detection algorithm based on Length Varied Average Magnitude Difference Function(LVAMDF) is presented. The new algorithm can be used to extract the average pitch period of the mandarin speech, when compared with the LVAMDF pitch detection algorithm. The result of the test experiments shows that the modified pitch detection algorithm brings good precision and better synthesized speech.With the above improvements and other speech feature extracting methods, the system of speech coding and speech decoding is realized. The result of the experiments shows that the proposed speech coder presents a good performance. The quality of the synthesized speech is still satisfying when the bit-rate of the coder is reduced to 2.35kb/s.