摘要
介绍说话人识别技术发展情况,阐述包括特征提取、识别算法和区分算法在内的文本无关说话人识别系统的整体技术框架和基本工作原理针对文本无关说话人识别相关技术给出了近几年主要发展的高斯超向量—支持向量机模型(GSV-SVM)、联合因子分析模型(JFA)和鉴别性向量(i-vector)模型,并对3种模型进行了分析比较:指出GSV-SVM模型可以提高识别系统性能;JFA模型能提高系统性能但计算量过大,难以实现应用;i-vector模型降低了计算量,并能提高识别精确度和效率,是目前的研究热点。最后指出当前文本无关说话人识别的研究难点和热点。
The development of the speaker recognition technology is introduced, and the overall technical framework in- cludes a feature extraction, recognition and differentiation algorithm; the text-independent speaker recognition system and the basic working principle are described. According to the given text independent speaker recognition technology in recent years, the main development Gauss super vector model of support vector machine（GSV-SVM） , analysis model of joint fac- tor（JFA） and differential vector （i-vector） model, and 3 kinds of models are analyzed and compared. We point out that the GSV-SVM model can improve the recognition performance of the system, JFA model can improve system performance but the calculation is too large, which is difficult to achieve application, while the i-vector model reduces the amount of cal- culation and can improve the recognition accuracy and efficiency, and it is the current research hotspot. Finally, we point out the current research difficulty and hotspot of the text-independent speaker recognition.
出处
《数字通信》
2013年第4期48-52,共5页
Digital Communications and Networks
基金
重庆市自然科学基金计划项目(cstc2012jjA40046)
重庆市公安局科研项目(2012-38)
关键词
文本无关
说话人识别
特征提取
模式识别
text-independent, speaker recognition, feature extraction, pattern recognition
