Rbf Neural Network Are Optimized Based On Genetic Voiceprint Recognition Research

Speech is to realize the communication between people is the most direct and the mostconvenient means, and how to realize the computer and between the smooth operator voicecommunications, has been to pursue a dream, speech recognition is the key technology ofdream. Along with the people to the identification and safety have become increasinglydemanding, biometric authentication technology for its unique advantages, increasinglyshows its value. And other biometric technology, voice recognition is not lost, no need toremember and easy to use. At the same time, voiceprint recognition also has its own uniqueadvantages, is the first to voice as recognition features, non contact and nature is theimportant reason for users to accept; secondly, voiceprint recognition using equipment costis very low, due to the telephone network and a microphone, computer integration, can saythe required hardware cost is almost zero; for remote applications and mobile Internetenvironment, at present, voiceprint recognition is likely to be the only solution, the speakerrecognition technology has become the people’s daily life and work of an important andpopular authentication mode.Although voiceprint recognition research for half a century, but the existing voicesystem there are still many problems, not up to the community for its practical requirements.In essence, the speaker recognition technology can be divided into two part featureextraction and recognition model. Therefore, in a sense, the root of the problem can bereduced by the feature extraction and recognition model limitations caused by. How to seeknew more characteristics of expressive force, has stronger robust speech feature, or to theexisting feature selection, optimization of fusion, compensation and other methods toenhance the performance of existing systems.This paper presents a RBF neural network optimization based on Genetic Algorithm invoice recognition algorithm, the genetic algorithm is used to the traditional RBF neuralnetwork RBF centers and widths are optimized, to overcome the traditional RBF neuralnetwork parameters are difficult to identify defects. At the same time, the algorithmcombines the psychoacoustic model, the performance of speaker characteristics extractionof Mel cepstrum coefficient as the feature for speaker recognition, can effectively improvethe system noise performance. The simulation results show that, compared with the traditional RBF neural network, the method has fast learning the weight of network capacity,and the network of the global optimization ability, enables the system to further improve therecognition rate.