基于量子逻辑线路神经网络的说话人识别方法  

Speaker Recognition Method Based on Quantum Logic Cir-cuit Neural Networks

作　　者：潘平[1] 罗辉[1] 王洋[1] 

机构地区：[1]贵州大学计算机科学与技术学院,贵阳550025

出　　处：《科技导报》2013年第33期15-18,共4页Science & Technology Review

基　　金：贵州省科学技术基金项目(黔科合J字[2012]2132);贵阳市科技计划项目(筑科合同[2011101]1-2号);贵州省国际科技合作计划基金资助项目([2009]700109;[2009]700125)

摘　　要：由于说话人的语音信号具有时变性、随机性,其特征参数也呈现出高维及相邻帧变化较大等特点。从量子信息处理理论出发,将一帧语音信号视为一个量子态,在传统神经网络的基础上,利用量子逻辑线路构造神经网络,实现说话人语音信号的有效聚类,探索一种基于量子逻辑线路神经网络的说话人识别模型与方法。利用模型固有的大量全局吸引子,可有效降低语音信号处理的时间及复杂度。通过在经典计算机上模拟仿真,并与BP神经网络说话人识别模型进行对比,表明该方法能够加快说话人识别模型的收敛速率,对参数变化具有更好的鲁棒性,且其系统识别率比BP神经网络方法平均提高了3.34%。Under the same spatial and temporal conditions, the quantum computing is superior to the traditional computing. Because a speaker＇s speech signal features the time-varying property and the randomness, its characteristic parameters also show high-dimensional characters and large changes in adjacent frames. This paper, based on the quantum information processing theory, takes a frame of the speech signal as a quantum state, and uses quantum logic gate circuits to construct the neural network according to the traditional neural network, and obtains an efficient clustering of the speaker＇s speech signal. A speaker recognition model is built and a method based on the quantum logic circuit neural network is proposed. This model has a large number of global attractors, and the method can use them to effectively reduce the complexity of the speech signal processing. Through simulations on a classical computer and a comparison with the BP neural network speaker recognition model, it is shown that this method not only can accelerate the convergence rate of the model but also has a better robustness with respect to the parameter changes. The system＇s recognition rate with the method proposed in this paper is 3.34% in average higher than that with the BP neural network method.

关 键 词：量子逻辑线路 神经网络 说话人识别 

分 类 号：TN912.34[电子电信—通信与信息系统]