基于HMM的声纹考勤算法改进  

The Improved Algorithm of Voiceprint Attendance Based HMM

作　　者：姚敏锋[1] YAOMin-feng(Cisco School of Informatics, Guangdong University of Foreign Studies,Guangzhou 510006, China)

机构地区：[1]广东外语外贸大学思科信息学院,广东广州510006

出　　处：《电脑与信息技术》2016年第5期22-25,共4页Computer and Information Technology

摘　　要：将声纹识别技术应用于签到系统,具有更为简便、准确、经济及可扩展性良好等众多优势。但在声纹考勤过程中,如果模型库中的模型数量过大,逐个匹配就会严重影响签到的实时性。文章研究改进的分组模型将具有相近声道特征的签到人分为同一组,在识别时只对同组模型进行计算,提高系统运行效率。同时为了弥补该种方法所牺牲的少量识别率,根据声音的物理特性,论文提出音调和音色的双特征HMM模型,提高系统识别率。The voiceprint recognition technology employs such advantage as simplicity, accuracy, cheap, good scalability and more when it is applied in attendance system. But during the voiceprint attendance checking on, if the number of models in the library is too large, matching one by one will seriously affect the real-time of checking. The improved grouping model that the paper investigated assigned the attending people have similar channel characteristics into the same group, when recognizing only calculate the model in the same group to improve the system efficiency. Meanwhile,in order to compensate the sacrificed tiny recognition rate that caused by the method, according to the physical characteristics of the voice, paper proposed double feature HMM model include tone and timbre and improve the system recognition rate.

关 键 词：HMM模型 分组 音调 音色 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]