The Research On Speech Emotion Recognition Based On Phase Space Reconstruction

With the rapid development of human-machine interaction, scholars generally agree that emotional information has been an important part of the cognitive process. And speech emotion recognition leads an important role in affective computing. Aiming at the limitation of the emotional characteristic parameters, the main work of this paper is shown as follow:According to the characteristics of the speech signal, we extract some usual speech feature, such as short-term energy (zero crossing rate, fundamental pitch) and spectral characteristics (MEL frequency cestrum coefficients). And the extracted data is computed to gain statistics. Then the statistics are regarded as original feature sets.Considering the complexity of the vocalization, this paper treats the speech signals as chaotic signals. First we extract lyapunov parameter from speech signals to judge whether the signals are chaotic or not. And take the phase space reconstruction theory into the sound emotional identification. By analyzing chaotic parameters of different speech emotion states, we extract correlation dimension and Kolmogorov entropy as emotion characteristic. The result shows that new parameters could present emotion information more comprehensively. Also new parameters can weaken the semantic of speech signals.In this proposal, we use SVM to recognize the state of emotion. Comparing traditional features with chaotic characteristics and the combination of all the characteristics, we find that introducing chaotic characteristics improve the performance of recognition. But it is easily confused under circumstance that sadness and peace occurs. Thus, an improved algorithm is proposed based on recognition of penalty factor and parameters of kernel function. By seeking for suitable parameters twice, emotion of sadness and peace are experimented respectively. And the experiment result indicates that recognition accuracy is improved significantly in the improved algorithm.