Research On Nonlinear Prediction Of Speech Signal Based On Volterra Model

With the increasing requirements of the voice quality and the extensive research on nonlinear theory and after the speech signal showed to have a very significant nonlinear characteristics,various nonlinear models of the speech signal had been established. Although compared to the traditional linear models,these nonlinear models have better prediction results.In order to further improve the prediction accuracy of the speech signal and to reduce the rate of the speech signal pass code,a new modeling approach which will let the speech signal processing technology to achieve a higher level must be found.Therefore,the nonlinear modeling of the speech signals has become a research focus of the speech signal processing.In this paper,on the basis of the speech signals samples with the chaotic characteristics,after using Volterra,the speech signal chaotic time series forecasting model has been established and the coding of the speech signals has be completed. For solving Volterra model parameters,the paper has presented a nonlinear regression algorithm.Through the application of the algorithm,the speech signal chaotic time series forecasting model with the adaptive capabilities has been got.Finally,through the experiments,the model was validated.The results show that the speech signal nonlinear regression proposed algorithm obtained chaotic time series forecasting model compared to the LPC model have higher prediction accuracy and lower pass code rate.The main works are as follows.1.The recognition of the speech signal chaos.The first step is the interception of a The speech signal time series and its sub-frame processing.Then Using the maximum Lyapunov exponent,the speech samples with chaotic characteristics were judged. Finally, using mutual information and Cao method of calculating the delay time and embedding dimension,the phase space of the speech signal has been reconstructed.2.The simulation experiment of the Lorenz time series has been completed.For Lorenz time series,the delay time and embedding dimension are obtained for phase space reconstruction.Using the nonlinear regression algorithm proposed in this paper,the Volterra prediction model has been obtained.Finally,the predicted results is simulated. 3.Using Volterra model,the process of the speech signal predictive modeling has been described.After the speech signal pre-processing and reconstruction phase space, using the Volterra model,each frame of data has been forecast,to achieve the speech signal coding.Simulation experiment of the speech signal has been completed.During the experiment,by comparison with the LPC model,its effectiveness has been verified in predictive modeling of speech signals.