Abstract: To efficiently detect SPIT calls in telecom network，and solve the contradiction between detection accuracy and efficiency，this paper proposes a new method combining filtering and exact recognition based on the analysis of SPIT characteristics.A simple decision based on VAD is firstly used to filter the suspected SPIT calls，then a SVM-based method is applied to detect the SPIT via speech feature finally.Experimental result shows that this method can efficiently detect SPIT calls with lower computation，which is helpful for real-time application.Moreover，the detection accuracy and real-time efficiency both increase while the speech length increasesing.

Key words: Spam Over Internet Telephony（SPIT）, Voice Activity Detector（VAD）, VAD series code, Support Vector Machine（SVM）

摘要： 为了对电信网中的垃圾网络电话（SPIT）进行有效的检测，解决该项技术应用中面临的检测精度和检测效率之间的矛盾，基于SPIT呼叫的典型特征，提出了一种粗过滤和精判别相结合的检测方法。基于语音激活检测（VAD）对疑似SPIT进行快速筛选，基于支持向量机通过语音特征对SPIT进行精确识别。仿真实验结果表明，该方法计算代价小、检测效果好、实时性高，并且随着检测语音长度的增加，该方法的检测精度和实时倍率都有所增长。

关键词: 垃圾网络电话, 语音激活检测, VAD序列码, 支持向量机