Research And Implementation Of Voiceprint Recognition System Based On Quantum Potential

Voice is not only the media communication between people,but also carry their own distinctive features.The voiceprint feature is a biological feature,which contains the speaker physiological,physical structure of different characteristics.It is not easily imitate and forged by others because of its high specificity.With the development of information technology,there are more and more civil and even criminal cases due to the characteristics of the voiceprint caused by identity forgery and some other primitive,authenticity relevant cases.It brings much trouble for judicial forensics and public security law enforcement.So it is of great significance to assist the public security handling cases and judicial forensics.It is of great significance to extract the characteristic parameters effectively for public security handling case and forensic evidence.In the speaker recognition system,the speaker feature extraction and feature matching are the two technical key points.The algorithm of speaker feature extraction will directly affect the effective extraction of the speaker's features.In this thesis,from the perspective of quantum mechanics theory,I try to research a new feature extraction method,which assuming the quantum barrier is a filter.Establishing barrier group via the correspondence between energy and frequency by using the tunneling effect of signal and quantum barrier.Quantify the transmission capacity of multi-frame speech with different energies.Treat the transmission probability when the speech frames pass through the barrier group.Transmission probability after trained by Gaussian mixture model is voiceprint features of different people.Finally,I programmed the voiceprint recognition system and tested the recognition rate.The system mainly functions are as follows: introducing and deriving the characteristic parameters of voiceprint,extracting and establishing the characteristic parameters and model of voiceprint,and comparing of voiceprint characteristic model.The purpose of the system is provide high-efficiency and high-accuracy of voiceprint matching auxiliary function for the police department and judiciary.The main work of the thesis is as follows:(1)First of all,I research that how to use the process-oriented voice signal processing and how to use the Windows API to achieve the voice signal inputting,preprocessing,detecting endpoint,framing and other theoretical techniques.(2)Then,I analyzed the same point with the MFCC algorithm from the theory of quantum barrier.And I tried to program this system.(3)Processing the parameters obtained after the feature extraction to ensure effective matching with the parameters required for the Gaussian mixture model.(4)In the system implementation,I have designed the various functional modules in detail and programmed the system.Finally,I carried out a small range of tests and find the influence level with different recording frames,different Gaussian mixing number on the recognition efficiency.The results of the comparative analysis show that when the number of effective frames(K)is fixed,the number of valid frames(training time)is higher and the recognition rate is higher.The higher the Gaussian mixture number(M),the higher the correct recognition rate,but the recognition time will increase.When K and M and other environmental conditions are unchanged,the more number of effective frames,the higher the correct recognition rate.