Speech Recognition Research Based On Deep Neural Networks

In the past few years, deep learning techniques, based on deep learning research,have an increasingly important impact on signal and information processing. It can besaid that deep learning has expanded to the main aspects of machine learning andartificial intelligence.This paper first introduced the basic principles of speech recognition, whichincludes pre-processing, MFCC features and speech recognition methods.As the main line of this paper, the application of artificial neural network onmachine learning was discussed. The deep neural network and deep learningtechnology were introduced. According to the differences of function and trainingmethods of the network structure, the deep neural network can be divided into threecategories: generative deep architectures, discriminative deep architectures and hybriddeep architectures. In terms of deep learning, the BP algorithm and theimplementation of deep learning technology by using RBM is introduced.Finally, Kaldi speech recognition toolkit, based on BP algorithm, is employed totrain deep neural network with4hidden layers and recognize Chinese speech.This paper research the influence of subsets of Chinese speech on the word errorrate (WER) of speech recognition realized by deep neural network.