A Self-Adaption Recognition Method for English Speech on Cloud Platform Based on Wavelet Neural Network

Abstract

Although there are various recognition methods for digital English speech (DES) on cloud platform network, the accuracy of them is low and the average time consumption is too long. Due to above shortages, this study put forward a self-adaption recognition method based on wavelet neural network (WNN). At first, the study used zero-crossing rate as feature of voiceless and voiced sound of the DES and used WNN to carry out conversion for the English speech information. Then, the study obtained wavelet-scaling function and parameterized wavelet function via zero-pole model. The parameterized wavelet function was used as the feature vector of the recognition. Wavelet function base of DES feature was generated via stretching and translating transformation. The transformed wavelet function was used to build WNN model. Moreover, error entropy function of the recognition was calculated by introducing momentum factor and partial derivative of the error entropy function to adjust parameter of the built WNN model. Thus, the study achieved the self-adaption recognition of DES. Simulation results show that the proposed method has high recognition accuracy and short recognition time.