The Emotion Recognition Of Speech Signal

Affective computing is developing very fast in the recent years.this research field mainly contain following parts.artificial affective, emotion recognize, emotion expression, etc. As the most important part emotion recognize is applied in many field, and is researched wildly. It concentrated embodiment of the modern science and the development of multi-disciplinary, multicross-cutting areas, mutual promotion and mutual infiltration characteristics, for example, facial expressions, voice, physiological signals, etc. This study is based on artificial neural network emotion and immune to emotional speech signal recognition.In this paper, introduce the affective computing research state first, and then elaborated on the artificial neural networks, artificial emotion theory, fractal theory, and so on.use the fractal value as the characteristic of the speech signal, then use the RBF to recognize.The paper also conducted the research the extaction of the speech signal featrue. These parameters mainly include pitch, formant, fractal, and so on. By using the self-related method, we can extract the pitch parameter, and extarct the formant by LPC, mesh compute for the fractal dimentsion. By the parameter we recognize different feelings. In the process of analysis, we found among the recognition of angry, happy, sadness, the difference of the emotion angry and happy is not obviouse, we must use a group composed of feature vector to reflect difference.The experimental results show that the using of fractal dimension to describe the chaos of speech signal and the using of RBF to identify are very effective. In the sample space, we can recognize which emotion the speech signal is in the rate of 100%. At the same time reflect RBF network convergence speed and the high rate of correct identification advantages. On the basis of RBF implementation, we train another network BP. Comparing the RBF with the BP. we found the convergence speed of BP is not very fast. Sometimes, it won't converge because of the incorrect initial parameters. RBF has the same functional approximation capability as the BP, and has the rapid convergence speed.The whole system is implemented by adapted the C/S construction. Using C++ to implement the frontier to accept the input of the client, using matlab to implement the back-part, extaction feature, training the network and recognition are its functions. The frontier communicates with the back-part through the MATLAB computing engine.