一种基于混合神经网络的说话人识别系统

目前说话人识别系统的识别率已经达到较高的水平,但是在应用系统的整合方面还存在一定的问题,比如其算法的复杂度高,不易于低成本的硬件实现等都限制其达到普及化.针对这样的问题,提出一种高性能、结构简单的基于自组织映射(SOFMNN)和概率神经网络(PNN)的混合神经网络分类器以取代目前常用的高斯混合模型(GMM)分类器.实验结果表明,混合神经网络分类器在识别率、识别速度和存储量上都优于高斯混合模型分类器.混合神经网络模型是一种高性能、高效率的说话人识别系统,该系统在说话人识别中将会有很好实用价值.

Speaker identification system based on hybrid neural network

At present,the accuracy of speaker identification is satisfactory,but some problems remain concerning system integration.For example,because of the complexity of the system algorithm,it is not easy to implement with low cost and thus it is not widely used.Proposed is a hybrid neural network classifier with high performance and simple structure that makes a substitution for the highly complex Gaussian mixture model(GMM).This classifier is composed of a self-organizing map neural network(SOFMNN) and a probabilistic neural network(PNN).According to experiment results,this hybrid network(SOFMNN-PNN) classifier presents better performance,higher calculating speed,and lower memory requirements than the GMM classifier.It is an effective,high performance speaker identification system with practical utility value.