Research On Intelligent Voice Print Recognition Method

With the rapid development of human society and economy and the comprehensive popularization of network communication and information technology,it is very important to verify personal identity information quickly,safely and accurately.Like fingerprints,palm prints,DNA,retinas,iris,and blood vessels,voice print is an inherent biological feature of everyone,with uniqueness and stability.In recent years,voiceprint recognition technology,which extracts the unique voiceprint information of each person to identify their identity,has attracted the attention and research of many scholars at home and abroad.The core of voice print recognition technology is to solve the preprocessing,feature extraction,pattern matching,judgment and scoring of the original voice signal.This paper mainly focuses on the above operation steps,the relevant research.There are still many factors affecting the recognition effect in the acquisition and processing of voiceprint information,such as the instability of voice signal,the easy loss of characteristic parameters of voice signal,the complex actual environment,the easy imitation of voice,and the insufficient sample parameters.Because these factors affect the reliability and accuracy of voiceprint recognition,there are still many key technical problems to be solved.Focusing on the structural features and influencing factors of the voice pattern recognition method,this paper proposes different solutions through MFCC(frequency cepstrum coefficient),CNN(convolutional neural network)and TL(transfer learning TL).MFCC and other methods can effectively solve the problem that voice signals are unstable and characteristic parameters are easy to be lost.It is necessary to pay attention to the selection and combination of different voice signals.The principle of MFCC is completed according to the auditory perception mechanism of human,which can effectively reflect the characteristic parameters of the speaker's speech.Moreover,this method does not need any presuppositions and has good recognition effect.Convolutional neural network(CNN)and other methods are effective in preventing speech imitation and speech denoising.Convolutional neural network is a matching method based on physiological pattern and a set of artificial neural network model simulating human brain.The model simulates the biological perception to a large extent,and has independent learning ability and robustness to incomplete information.It can distinguish the complex classification boundary effectively,and it is a neural network model of distributed parallel processing structure,whose performance is very close to the ideal classifier.The migration learning TL method can effectively solve the shortcomings of insufficient samples of convolutional neural network and the inability of small samples to make the network converge.Transfer learning is a new method of machine learning,which mainly uses existing knowledge to solve problems in different but related fields.Transfer learning is a method that researchers use existing knowledge to solve similar problems accurately and quickly.Since the mid-1990 s,it has been favored by more and more scholars.Finally,experimental comparison and verification of the three methods are carried out in this paper,and it is proved that the combination of the three methods can effectively improve the voiceprint recognition rate.This paper analyzes and summarizes the experimental results,and gives a detailed description of the shortcomings in the experiment,and predicts the development direction of voice print recognition technology in the future.