Research on multi-base depth neural network speech recognition

Abstract

In speech recognition system, an improved multi-base neural network speech recognition model is proposed to solve the problem of long learning time and slow convergence rate of deep neural network. However, the improved model introduces a large number of parameters in the training process to make the model over-fitted in the test set, resulting in the deterioration of generalization ability and the decrease of speech recognition rate. In order to solve this problem, an improved BP algorithm is introduced in the last layer of the depth neural network model. The experimental results show that the multi-base neural network model not only reduces the training time and calculation amount, but also improves the recognition accuracy compared with other recognition models, and it is an excellent speech recognition model.