Research On A Linear Predictive Coding And Decoding Algorithms

In the communication system, the bandwidth is a very valuable resource. How to obtain a higher quality voice and a better user experience is the low-rate speech coding problem to be solved under the limited bandwidth. Its goal is to provide better voice service in the lower transmission rate, or to take up minimal bandwidth resources at the same quality premise. Speech coding can transmit analog signals into digital signals. Usually, speech coding can be divided into the following three categories: waveform coding, parameter coding and hybrid coding. Waveform encoding speech signals are processed as a normal waveform signal, the speech signal is reconstructed, is similar to the original speech signal. Parameters encoding extracts speech feature parameters, and these parameters are encoded. Hybrid coding combines the advantages of both waveform and parameter coding. In this thesis we use hybrid coding to design vocoder.Aiming at the existing problems, complex parameter extraction algorithm and low synthetic speech naturalness. We make a series of optimization improvements. The main work includes several aspects: 1) use mixed incentives to replace simple binary incentives;2) refined excitation signal to simplify the pitch extraction algorithm;3) improved real-time voice. This paper presents a fuzzy clustering secondary cascade LBG codebook training algorithm, combines fuzzy clustering and two-stage codebook their strengths and uses a secondary vector quantization algorithm proposed sub-band sound intensity vector quantization.Finally, we do simulation experiments about the proposed algorithm and compare with the traditional method. Experimental results show that the proposed algorithm is excellent at the time complexity of the algorithm, synthetic speech quality and other aspects of performance.