Nonlinear Feature Etraction of Eotional Speech Based on Caotic Caracteristics

Based on the chaotic characteristics of speech utterance and propagation, the experiment firstly verifies that speech signals with different emotions have chaotic characteristics. Secondly, the experiment tries to adopt the nonlinear dynamics theory to extract three kinds of nonlinear characteristics which are based emotional speech signal. The characteristics include the minimum time delay, correlation dimension and Lyapunov exponent. Thirdly, designed different experiments to verify the recognition performance of the non-linear characteristics. Furthermore, the experiments have selected emotional statement from Berlin speech database and have used a support vector machine to identify the emotion. And the parameters are obtained by ten-fold cross-validation. Finally, through the summarizes and analyzes of the experimental results from different experiments. Then it compares the relative merits of different nonlinear characteristics which were used to identify the basic emotions.