Speech Emotion Recognition Method Research Based On RBF

Speech emotion recognition is an important research subject in the field of emotion computing, which is a premise for realizing intelligent human-computer interaction. Speech emotion recognition research is in an initial stage, how to improve the recognition speed and the rate of recognition has become a difficult problem that can not be ignored and need to solve.The feature parameters which are exacted from traditional voice emotion recognition include pitch parameters, short time energy, amplitude , Fourier transform spectrum signal characteristics, speed, etc, to find emotional features that reflect the parameters of speech in order to improve speech emotion recognition rate and recognition speed. Firstly this paper researches on extraction feature parameters, choosing the prosodic features of the speech signals and sound quality characteristics as the characteristic parameters of emotion recognition, proposes joining the fractal dimension as the new feature parameters of the speech signal, fractal dimension is extracted by using the box dimension calculating method. According to happiness, anger, sadness, fear and neutral five emotional speech signals, this paper analyzes the prosodic features of speech signals and voice quality features, finds distribution laws of different emotional speech feature parameters.This paper competitive learning mechanism, gradient descent and the deletion policy combining algorithm, obtained the right hidden layer nodes, the cluster center and radius. Also used the gradient descent method of training makes the network has a faster speed optimization. RBF neural network is proposed based on the establishment of a speech emotion recognition system, and also trained a BP network and probabilistic neural networks compared test results from the experiment can be seen that RBF neural network based on emotion recognition in the recognition rate and recognition have a significant speed increase.