Research On Identity Recognition Algorithm Based On Speech Features

The realization of identity recognition based on voice signals is the focus of current research.It has been widely used in password verification,voice control,military communications and other fields.Compared with other identification technologies,the use of voice signals for identity recognition has the advantages of easy signal acquisition and safer identification.This paper focuses on the key technologies and algorithms involved in the realization of identity recognition based on voice signals,and carries out the following research work:The paper first introduces the research background and significance of the speaker recognition in detail.Then,this paper systematically sorts out the research status and key research methods of each stage,summarizes the main research ideas,main algorithms and key technologies of speech identity recognition technology,and deeply analyzes the basis of the basic principle and main technical route of speaker recognition,and analyzes the problems in the current research technology.Secondly,the high-quality voice signal acquisition method and the front-end preprocessing technology of the voice signal are studied.The time domain characteristics of the voice signal including short-term energy and short-term average amplitude are mainly analyzed.The Mel cepstrum coefficients in frequency domain features are solved.At the same time,the speech signal enhancement technology based on adaptive filter algorithm and spectrum reduction algorithm is studied.Meanwhile,the characteristics of spectrogram and its calculation method are analyzed in detail.Again,this article discusses the feasibility of deep convolutional neural networks and recurrent neural networks in the speaker recognition's application scenarios deeply.By using the spectrogram as the input signal,this paper constructs a speaker recognition network model based on deep learning algorithm.On this basis,the design process of CNN layer and LSTM layer in deep learning network is analyzed in detail.Then,this paper uses the two scales of spectrogram to determine the front-end deep CNN structure of the deep learning model,and it adds recurrent neural network and Gaussian noise layer to construct a network structure with 5-layer CNN layer + regularization layer + 3 layers.The new deep learning network structure has been verified the accuracy of model recognition with tested on the self-built speech data.The validity of the designed deep learning network model is verified by combining the self-built data sets with high and low voice and different speech duration.In the end,this paper developed a fully functional voice identity recognition application software based on the PyQt framework,and designed the composition,working principle and algorithm flow of the identity recognition application software in detail.At the same time,based on the self-developed voice identification recognition application software,a large number of experiments and algorithm verification work were carried out to verify the effectiveness and accuracy of the algorithm studied in the paper.