Research On KK-SOVF Speech Signal Sequence Prediction Model Based On Particle Swarm Optimization

Recently,voice signal processing is an important research topic field and it can be the foundation of secret communication of signal processing, digital communication, voice recognition.As the voice signal was proved to have obvious nonlinear characteristics, many various nonlinear models had to start to establish. Although compared with the traditional linear model, the nonlinear model has better prediction effect. But in order to further improve the prediction precision of the voice signal, we must use the new methods which will help us to achieve a higher level. So the research of the nonlinear modeling method had become the improtant study of the voice modeling study. And the research of the nonlinear modeling method also can provide more methods for voice signal processing. In this thesis, using the second order Volterra which has forecast modeling research for the model of voice signal with time series.model adopted is easy to implement and has the global search ability of uniform search particle swarm optimization algorithm for on the model parameters to establish the prediction model of voice signal display structure. On this thesis, the research on the constructed within the scope of permissible error UPSO-SOVF model had been streamlined which help us get UPSO-KK-SOVF contracted prediction model. In this thesis, the main work are as follows:First, distinguish the voice signal has identification. First using the sequence and carries on the frame to extend the interception of a voice signal time, and then use the use the method CAO which was improved to calculate the voice signal and Cao method is used to calculate the embedding dimension, the reached the delay time and embedding dimension can carry on the phase space reconstruction as a key parameter, finally USES the maximum Lyapunov index to judge whether the voice signal has chaos characteristics.Second, UPSO-SOVF speech signal prediction model is constructed in this thesis. After the voice signal is used to identify the chaos, then using the model Volterra which was based on UPSO modeling to project the voice and data,then we can use it at the prediction of voice signal. UPSO has faster convergence speed, and it also has obvious advantages in getting rid of local extreme value,and improving the convergence precision. The simulation results show that UPSO-SOVF compared with LMS-SOVF model, UPSO-SOVF predicts more accurate than LMS-SOVF model, thus we can verify the effectiveness of the model in the model voice signal prediction. Using the second order model Volterra on the model parameters as the particle position, as the column, the whole second order model Volterra as the fitness function, thus solving the optimal model parameters, the corresponding column effectively.Third, UPSO-KK-SOVF speech signal prediction model is constructed in this thesis. In this thesis, to lean on the model on the premise of setting allows errors, according to different experimental data and the permissible error of the effect on the model of the column selection is different, but the number of valid column is far less than the number of the original on the model. The experiment proved that the model is a redundant items, we can lean on the model by setting the allowed error range for certain. In order to prove UPSO-KK-SOVF method is effective,we can base on the root mean square error and waveform comparison which is through experiment simulation.