Research On Speech Perceptual Hashing Authentication Method And Its Application In Mobile Terminal

Nowadays,some subjects,such as computer science,pattern recognition,artificial intelligence and so on,have been achieved a remarkably progress.Based on them,speech content authentication technologies gain a rapid development.As the common multimedia information,speech is susceptible to diverse kinds of attacks.With the popularization of mobile device,human need more advanced speech content authentication technology.In order to implement more efficient speech authentication process,improve its anti-interference ability and apply it to mobile clients,this thesis is mainly researching some speech perceptual hashing authentication algorithms which can be used to mobile terminal and achieving a speech identity recognition APP.Specific contents are as follows:1.In order to meet the requirements of discrimination,robustness and high-efficiency authentication of existing speech authentication algorithms in mobile speech communication,an efficient perceptual hashing scheme based on spectrogram for speech authentication was proposed in this paper.Firstly,a fraction of spectrogram is cut,which represents low frequency information of input speech signal and is less susceptible to common content-preserving manipulations such as MP3 compression,noise addition and volume adjustment etc.Secondly,the local binary pattern(LBP)algorithm is applied to produce a LBP feature image.Finally,the perceptual hashing sequence is obtained by employing an image perceptual hashing algorithm to the LBP feature image.Experimental results show that the proposed approach has a good discrimination,robustness and authentication efficiency.It can satisfy the real-time authentication requirements in mobile speech communication.2.In order to get strong robustness in real-noise environment,a robust speech authentication algorithm based on quasi-harmonic model(QHM)was proposed in this paper.Firstly,the speech signal to be authenticated need to be done some preprocessing,including filtered,windowed and framed.Secondly,each frame signal is used to compute its QHM on the condition of inputting some reference frequency,and then each of them is corresponding to a single-component modeling analysis signal.Finally,the prominent amplitude information coming from modeling analysis signal of each frame is extracted as perceptual features,which is used to generate binary hashing sequence and complete speech authentication.Experimental results show that the proposed algorithm has excellent robustness to real-noise addition and common content preserving manipulations,meanwhile its discrimination and authentication efficiency can meet authentication requirement of mobile speech communication.3.In order to apply existing speech content authentication algorithm to mobile terminal,in this literature,an identity recognition APP based on speech perceputual hashing algorithm is designed on Android platform.The design programs of its client and server are introduced.Firstly,an Android client is designed to offer user interface and record user voice.Next,a speech perceptual hashing algorithm is running on Tomcat server to capture and store user information.Lastly user client and server are communicating with each other to attain the purpose of user indentity recognition.Through deployment and test of entire system,experimental results show that the design module of speech authentication APP that is used to register,login,send or receive information,record speech can be ran smoothly and the performance of the system can meet predicted design requirements.