Speech recognition based on denoising self coding neural network
Qin tao, Hao Dong, Yang Yong, Yifei Zhang, Yang changmao, Yuxuan Liu · 2020
With the rapid development of artificial intelligence, speech signal is one of the main signal sources of human identity recognition. Improving the accuracy of speech feature signal is an important work to improve the rate of identity recognition. The neural network algorithm based on de-noising and self coding can effectively remove the noise, make the model trained by the recognition network better, and improve the accuracy of the recognition network for speech identification. The MFCC algorithm is used to extract the features of multiple speech samples. The denoising and self coding neural network training is carried out for multiple feature samples to minimize the reconstruction error between the trained output feature samples and the original samples. Neural network is used to train the model of self coding feature samples, and to recognize the speech identity of the test samples. Under the interference of multiple noisy speech, the recognition rate of speech identity is higher than that of traditional recognition methods.