基于网络的说话人识别系统研究-信号与信息处理专业论文.docxVIP

基于网 基于网络的说话人识别系统研究 I万方数据 I 万方数据 摘要 语音是个人所固有的特征，具有唯一性，用语音来鉴别说话人的身份也因其独特 的优点而日益受到人们的关注。在互连网飞速发展的今天，如果能够将说话人识别技 术和网络传输技术有机地结合起来，通过因特网对个人身份进行远程识别（尤其是与 文本无关的识别），对于人工智能的发展必将具有极其深远的意义。 本文将基于网络的与文本无关的说话人识别系统作为研究及设计的对象，针对客 户端内存容量有限以及网络传输的特点，在 Visual C＋＋6.0 软件环境下设计出了适 合于网络环境的说话人识别系统。该系统采用客户/服务器模式，以 Windows Sockets 类构建的 TCP/IP 协议进行网络连接，负责客户端特征序列的传送和服务器端识别结 果的回传。在客户端进行特征提取，将传统的线性特征与一种新的非线性特征结合， 得到一组数据量小、特征覆盖面高的说话人特征序列；在服务器端进行识别，将隐马 尔可夫模型（HMM）与支持向量机（SVM）相结合以满足识别率高、实时性好的系统要 求。另外，为了降低数据量、提高系统识别效率，在端点检测和识别方式中还分别采 用了改进的对数能量形式和预分类的识别方法。实验结果显示，该系统最终识别率达 到了 90％以上，具有较好的实用价值。 本文最后给出了研究结论，指出了不足之处及未来的研究方向。 关键词：说话人识别；客户/服务器模式；TCP/IP 协议；隐马尔可夫模型（HMM）； 支持向量机（SVM） II万方数据 II 万方数据 Abstract Nowadays, to identify speakers with voice is attracting more and more attention considering voice is an inherent and unique feature of every one. It will contribute greatly to the development of artificial intelligence to combine speaker recognition with the technology of network transmission and to recognise speakers long-distance (especially text-independent speaker recognition). The system of text-independent speaker recognition based on network is discussed in this paper. Considering the limited capability of the Client and the demand of network transmission, a system which is appropriate for network environment is achieved with Visual C＋＋6.0. The Client/Server model is applied in the system and the TCP/IP protocol which is constructed by Windows Sockets classes is used as network connection to transmit feature parameters from the Client and carry back the recognition result from the Server. Feature parameters is extracted by the Client, and then a group of speaker vocal features is formed by combination of traditional linear features and a kind of new non-linear feature in order to decrease data quantity and include more features. The process of recognition is done in the Server , HMM is combined with SVM to improve the rate of recognition and to response in real time. In addition, an improved