Research On Voiceprint Recognition Algorithm Based On Integrated Learning

In recent years,with the development of science and technology and the rise of artificial intelligence,biometric identification technology has become more and more important in various authentication scenarios.Voiceprint is a unique biological feature of an individual and is widely used in many fields such as military security,forensic identification,medical and health care and criminal investigation with its advantages of directness,convenience,flexibility and economy,and is gradually developing into the mainstream trend in the field of biometric technology.Integrated learning method has advantages in accuracy,stability and generalization,and is widely used to solve various problems.At present,it has made some achievements in voiceprint recognition system.However,the traditional integrated learning methods may not achieve satisfactory performance when dealing with complex data,such as changeable environment,noise interference,and unbalanced sample parameters,etc.Moreover,integrated learning is dedicated to the study of a specific scenario,and the characteristics such as diversity,accuracy and generalization contained in the model contradict each other to improve the performance of the model in some cases,so it is not suitable for various learning scenarios.In this paper,a new immune integration algorithm based on artificial immune knowledge is proposed,which can quickly and optimally classify voiceprint data in various learning scenarios.The innovation and research work of this paper mainly include the following points:(1)Aiming at the problem of low recognition accuracy of traditional methods,the Immune-Adaboost algorithm is proposed and applied to gender-specific voiceprint recognition.The algorithm solves the problems of complex algorithm and large training error of existing ensemble learning algorithms by using the biological immune mechanism and the synergistic effect of immune models,and performs well in voiceprint recognition.The results show that the Immune Adaboost algorithm has a high recognition accuracy in solving the problem of gender voiceprint recognition,and the effectiveness of the Immune-Adaboost algorithm is fully verified using the publicly available UCI data and voiceprint data.(2)Aiming at the problems of unbalanced recognition effect and low accuracy faced by traditional methods in dealing with unbalanced data,the Neuroimmune-Adaboost algorithm is proposed and applied to voiceprint recognition of unbalanced data.Inspired by the model evaluation criteria,the method adds the misjudgment cost factor(c_i)and the misjudgment weight adjustment factor β_i)provided by the neural system model on the basis of Immune-Adaboost,which greatly alleviates the problem of unbalanced data recognition.The model is verified by disclosing common unbalanced datasets and imbalanced voiceprint data.The results show that the Neuroimmune-Adaboost model has better recognition effect on both kinds of datasets and can achieve optimal recognition.(3)Aiming at the poor robustness and running speed of traditional methods in dealing with noise data,the Complement-Immune-Adaboost algorithm is proposed and applied in noise voiceprint recognition.The algorithm improves the decision-making strategy of the innate immune model by introducing the complement system,and puts forward the Complement-Immune-innate complement immune model,which further improves the effectiveness of the Complement-Immune-Adaptive complement immune model.The proposed algorithm has better recognition accuracy when dealing with the problem of noise voiceprint recognition.In the verification experiment,Gaussian data set,public UCI data set and noise voiceprint data are also used to verify the robustness and rapidity of the Complement-Immune-Adaboost classification algorithm.In this paper,the proposed algorithm is mainly compared and validated with Adaboost,LogitBoost,RobustBoost,LPBoost,GentleBoost,KNN and other traditional algorithms,and the experimental results show that the proposed algorithm has the best recognition effect.Furthermore,the effectiveness of the algorithm is further verified by quantitative indicators such as Accruacy,Recall,F1-Measure,Precision and Specificity.The comprehensive experimental results show that the immune ensemble learning model proposed in this paper has a broad application prospect in pattern recognition and speech recognition.