Speech Recognition Based On RBF Neural Network

Speech recognition has received more and more attention recently due to the important theoretical meaning and practical value. Up to now, most speech recognition is based on conventional linear system theory. With the deep study of speech recognition, nonlinear system theory method must be introduced. From the nineties of the twentieth century, with the development of nonlinear-system theories such as artificial neural networks (ANN), it is possible to apply these theories to speech recognition. Radial Basic Function Neural Network (RBFNN) offers a novel and effective way for studying as a feed forward multi-layered network, its study and application has developed rapidly Recently.This paper mainly studies speech recognition based on RBFNN. Computing validation performance analysis and results assessing are handled to each part of speech recognition process such as preprocessing, feature extraction and recognition algorithms.A RBFNN is a kind of feed forward network which basically involves 3 layers. It is proved that its speed of convergence is much quicker than general Back Propagation (BP) algorithm, and its structure can be fixed in the algorithm. The main problems in designing a RBFNN depend on fixing the nodes of the hidden layer, the parameters of the centers and the linear weights. This paper uses this method to construct RBFNN: Uses the combine of competition and clustering arithmetic to train the parameters of the hidden layer dynamically; Gradient-descent is used to train the weights which make the cost function minimized; To save the resource, Akaike's Final Frediction Error (FPE) standard is employed to delete the nodes that contribute little to the outputs of the network. This will balance the precision with the complexity of the network. Until the value of FPE no longer drops, a group of final optimum weights and a rational network have been found. Otherwise, this paper constructs RBFNN with Iterative method, Select fixed center random and Probabilistic Neural Networks (PNN).This paper uses VAD to detect speech point, uses Mel-Frequency Cepstrurn Coefficients (MFCC) to get speech characteristic parameter, uses Dynamic Time Warping (DTW) to adjust parameter finally. Then train the RBFNN with the train data. Finally input the test data to the learned network to recognize, recognition result based on pure speech and noisy speech samples shows that this improved RBFNN achieves excellent performance in terms of recognition rate and recognition speed.