Research On Identity Authentication And Anti-Spoofing Algorithm Algorithm Based On Voice Print Recognition

Biometry-based identity authentication method has gradually become an important method in identity authentication technology by virtue of its ability to represent identity,its convenience of use and its advantages of not easy to steal.Among the biometric features,voice print contains not only the physiological characteristics of individuals,but also the behavioral characteristics of personalities.Compared with the static features such as fingerprints and faces,it is more secure,which is the research focus and future trend of identity authentication technology.At present,the bottleneck of voiceprint recognition technology in practical application is its recognition stability in the face of changing environment and security capability in the face of fake voice attack.Firstly,in the practical application,in addition to the interference of ambient noise,changes in the speaker's own state,such as illness and mood fluctuations,will change the sound characteristics,thus affecting the stability of voice print authentication.In the face of this problem,the current voice print recognition technology has gradually developed from single-feature modeling to multi-feature identity modeling,aiming to extract more stable identity information from multi-feature.However,most of the current multi-feature fusion methods are direct splicing,which cannot effectively play the role of various features,and will greatly increase the feature dimension and training complexity.Secondly,with the development of technology,some forged voices are similar to real ones,which greatly threaten the security of voice print authentication technology.In view of this problem,anti-spoofing algorithm has been specially researched in the field of voice print recognition,but the current countermeasures are mainly to discuss all kinds of attacks separately,without effective fusion of voice print verification tasks,which limits the practical application of voice print identity authentication technology.Therefore,this paper carries out three tasks for the above problems.First,it combs all kinds of voiceprint features,conducts experiments on the effectiveness of all kinds of voiceprint features respectively based on the ideal environment data set and the abnormal voice data set,and compares and analyzes the stability of all kinds of features based on the experimental results.Then,based on the experimental results,the multi-layer voiceprint feature set is constructed in this paper,and the two-layer multi-feature fusion algorithm based on Fisher ratio is proposed.The fusion mechanism is based on Fisher ratio to perform in-layer multi-feature fusion and hierarchical decision weighted fusion of multi-layer features from the feature layer and the decision layer respectively.Results show that the multi-feature fusion model presented in this paper performs better than the single-feature model.Finally,this paper proposes a voiceprint verification and anti-spoofing algorithm based on multi-task learning and multi-feature fusion.The algorithm combines the Fisher ratio based multi-feature fusion method and the multi-task learning mechanism to model the voice print verification and anti-spoofing algorithm simultaneously.The two tasks complement each other so that the voice print verification task contains defense information against spoofed speech.Experiments show that the model can improve the stability and security of voiceprint verification.