The Research Of The Speech Encoding & Decoding In The Digital Processing Of Speech Singnals

In the digital processing of speech signals,speech coding technique is most important.In the era of digital information,the application of speech coding is very abroad.So it is necessary to study perfect speech coding arithmetic in the digital processing of speech signals.In this paper,the basic theories in the digital processing of speech signals are outlined,the speech coding and encoding arithmetics are discussed,and the linear predict coding is improved.In the improvement of speech coding,the neural network is applicated in the vector quantization .In the processing of amending the linear prediction coding,the quality of regenerative speech is improved by amending the inputted speech source to mixed speech source. Such as adding two impulses into the Gauss noise in the input speech source of surd or adding Gauss noise into the periodic impulses in the input speech source of sonant.In this paper,the degree of the comparability between the LPC coefficients is scaled by relativity,thus the transmission speed of LPC coefficients is debased.And the vector quantization based on classification is applicated to reduce the speed of speech coding and to ensure the quality of the regenative speech.The producing of speech signals is a non-linear process,speech signals are non-linear.In this paper,the non-linear prediction of coding arithmetic is discussed basing on the neural network arithmetic.Compared with the tradictional linear prediction of speech signals systems,the systems based on the neural network replace the models of linear prediction coding with neural network.By this means,the complexity of the arithmetic can be decreased and the SNR of speech coding arithmetic can be increased.Vector quantization is an efficient technology of compressing data and it is applied widely in the regions of speech singnals coding.The vector quantilizer based on neural network deals with the input . vectors by parallel way,the coding book is storaged in the connect weighted vectors dispersedly.So the searching of coding book can be much faster and more efficient. In this paper,the result of vector quantization can be better by applying the neural network arithmetic.In the realization of these kinds of mended arithmetics,the TMS320C5402 DSK of TI company is adopted as the hardware flat roof and the CCS integrated exploiture environment.And the software of the whole system can be optimized to improve the efficiency of programs.