Speech Recognition Algorithm Based on Deep Convolutional Neural Network

Language and speech are the most natural, convenient and effective means for human to obtain information. Speech recognition enables the machine to extract the information contained in the speech, which is equivalent to providing the hearing ability for the computer system, which can help people and machines achieve a more natural interaction experience. With the continuous enrichment and expansion of speech recognition theory, a variety of new speech recognition equipment has been put into use. DNN model cannot process massive speech data, and the performance of speech recognition system is directly affected. By improving and optimizing the DCNN model, the complexity of DCNN can be controlled and the level of speech recognition can be improved. In this paper, two kinds of DCNN models, DCNN-SR and DCNN-RR, are constructed based on the basic speech model and CNN model, and a new training method is studied. The simulation experiments of baseline CNN and DCNN models are carried out by selecting the CTS-30 speech data set, and the advantages of the constructed models in the field of speech recognition are verified.