基于支持向量机的关键词拒识算法

Keyword Rejection Based on Support Vector Machine

支持向量机是统计理论学习中一个重要的学习方法,也是解决模式识别问题的强有力工具,尤其在二元分类上有着突出的优势。拒识技术是语音识别系统走向实用化的关键技术之一,但由于语音信号的复杂性,使得拒识一直是语音识别技术中的难题。有机的将支持向量机技术应用于关键词识别的拒识问题中,把关键词识别中的正识和误识作为支持向量机的二元分类对象。这种方法避免了传统拒识方法对拒识门限的确定,同时充分发挥了支持向量机在二元分类上的优势,实验表明该算法效果较为有效。

Support vector machine is an important method of statistics theory, which is also a powerful tool in solving pattern recognition questions. Especially its advantage in clustering two classes. The technology of rejection is one of the most crucial technology in speech recognition marching practical applications, however, which is the most difficult questions in speech recognition because of the complexity of speech. The article applies the technology of support vector machine to the keyword rejection, which regards the accurate recognition and false recognition as the object of clustering. This method avoids selecting the rejection threshold in the traditional rejection method, at the same time, which makes good use of the advantages of support vector machine in clustering two classes. Finally, the experiment shows the efficiency of this rejection method.