Research And Implementation Of Home Equipment Control System Based On Speech Recognition

With the continuous development of science and the popularization of smart terminals,a new understanding of lifestyle is appeared.A comfortable,smart,and convenient home environment is becoming mainstream in modern society.Therefore,the smart home industry rises rapidly.As a common command control method,the speech recognition control technology is becoming more and more widely used in smart home.The home equipment control system based on speech recognition gradually become the focus in the field of artificial intelligence.In this thesis,the home equipment control system based on speech recognition is deeply studied,the main works are given as follows:I.The stratified self-adapting denoising algorithm in the home noise environment is researched.Firstly,this thesis studies the classification of noise under different standards.Secondly,the principles of several traditional denoising methods are recommended,and their advantages and disadvantages according to the denoising results are analyzed.Finally,the stratified self-adapting denoising algorithm is proposed and verified,by adding mechanical noise and artificial noise to pure voice instructions to simulate the home noise environment.II.The combinational neural network speech recognition algorithm in the home speaking environment is researched.Firstly,this thesis studies the model structure and training algorithm of Deep Neural Network(DNN)and Long Short-term Memory Network(LSTM).Secondly,according to the characteristics that LSTM can record long historical information by using the memory unit,while DNN can effectively extract high-level information in the data,a theory of adding LSTM to the first layer of DNN's hidden layer is proposed,and a combined DNN-LSTM network is built.Thirdly,according to training acoustic model by using the voice data set for different languages in combined models,a process of language matching is proposed by introducing information entropy.And a best result is produced by comparing the information entropy of two acoustic models' output probability.Finally,the performance of the DNN-LSTM theory mentioned and the effectiveness of language matching process are verified.III.The home equipment control system based on speech recognition is designed and implemented.Firstly,this thesis describes the software functions and hardware components of the home equipment control system based on speech recognition.Secondly,the functional modules of the system are divided and introduced in detail.Finally,the system is tested in terms of the recognition rate and recognition speed,and the advantages of the system are verified by the test results.