Research On Speech Signal Prediction

Speech communication plays an important role in people’s life, the real-time of the speech signal is starting to be more widely appreciated. But because the restriction of the channel bandwidth, the speech signal must queue in the transmission process, causes the blocking phenomenon of information, leads to the voice delay, voice interruption and so on. So the voice signals must be “slimming” to reduce the symbol rate, which can make the speech signals transmit in high quality.The problem that much data of the speech signals has to be transmitted was analyzed, the theory of voice predictive coding was studied mainly, the vector quantitative technology and the design of codebook were researched deeply. Traditional voice predictive coding algorithm use the linear prediction coefficients instead of the voice itself only, the data was still huge. This paper presents an algorithm of transmitting linear predictive coefficients with the codebook, the appropriate predictive coefficients were selected as vectors in codebook, The vector quantization was applied to linear prediction coefficients algorithm was proposed, used the vectors in the code-book instead of the obtained linear prediction coefficients, further compression of the speech signals by transmitting the vector code was achieved. The results of the MATLAB proved that the improved algorithm compressed the transmitted data of the speech signals greatly.The paper researched the method of vectors’ determine in the codebook. Two generation methods of the code-book were proposed: probability method and the classification based on the linear predictive coefficients. The results of simulation proved that, the recovered speech signals can satisfy the peoples’ hearing by using the reconstructed coefficients that quantified by the two code-books.