Research On Kernel Extreme Learning Machine Optimized By Grey Wolf Optimizer Algorithm For Speaker Recognition

In today's society,the biometric authentication technology based on the extraction of human features continues to develop.Speaker recognition,as one of them,distinguishes the identity through the voice of the speaker.This biometric authentication technology is also called voiceprint recognition,and its core content is to process the information that can represent the identity of the speaker extracted from the voice sample signal to realize the recognition of various types of speakers.As biometric authentication methods have made great progress under the promotion of machine learning in recent years,they have broad development prospects and important research value.Speaker recognition is also deeply affected by it.Scholars have gradually changed from the research of speaker recognition based on traditional methods.The direction shifted to speaker recognition based on machine learning methods.This paper proposes a method of applying the kernel extreme learning machine optimized by the grey wolf algorithm to speaker recognition.By selecting a suitable speech sample data set,extracting the voice features in the data set and reducing the dimensionality,the reduced data set Put it in the core extreme learning machine model optimized by the grey wolf algorithm for training.Finally,the test sample is used as the input of the system after training,and the matching calculation and result comparison are performed.First,collect voice sample data.Select 15 experimenters to collect 30-40 pieces of voice data per person in a quiet environment,and process the collected voice sample data into WAV format files of about 9 seconds.,In order to reduce the complexity of the speaker recognition system,the obtained data is subjected to feature processing,which provides real experimental data for speaker identification.Secondly,the feature-processed speech sample data set is divided into two categories,one is used as a training set for the learning of the speaker recognition system,and the other is used as a test set to compare the recognition results of the speaker's identity.After proposing the extreme learning machine model,the kernel extreme learning machine model was studied in detail through formula derivation and theoretical analysis.By identifying a set of random sample data,and selecting different traditional machine learning methods in the identification,comparison experiments were carried out.It shows that the training speed and accuracy of the nuclear extreme learning machine are far superior to traditional machine learning methods.Finally,in order to pursue better recognition accuracy,the grey wolf optimization algorithm in the swarm intelligence optimization algorithm is introduced.The experimental results show that the core extreme learning machine optimized based on the grey wolf algorithm is superior to other swarm intelligence in terms of recognition accuracy and convergence speed.The network model is optimized by the algorithm,and the optimal solution is sought in the iterative optimization of the grey wolf algorithm,so that the maximum recognition accuracy rate for speaker recognition in an experiment is 94.76%,and the average recognition accuracy rate is 93.17%,and each category of speakers The accuracy of voice recognition is above 85%.The results show that the improved network model has better recognition accuracy than the nuclear extreme machine network model,and that the nuclear extreme learning machine network model optimized based on the grey wolf algorithm is better.The classification characteristics of this paper verify the good development prospects of the nuclear extreme learning machine network model based on the grey wolf algorithm optimization in the speaker recognition application.