基于DSP的小型语音身份验证系统  

Small-Sized Voice Identity Recognition System Based on DSP

作　　者：赵振东[1] 齐苗苗[1] 周娜 肖学东[3] 

机构地区：[1]华北电力大学电子与通信工程系,河北保定071003 [2]河北省农村信用合作社网络中心,河北石家庄050021 [3]黑龙江省电力有限公司通信自动化中心,黑龙江哈尔滨150090

出　　处：《现代建筑电气》2010年第8期15-18,37,共5页Modern Architecture Electric

摘　　要：为了将说话人识别技术实用化,用DSP建立了一个闭集小词汇量、与文本有关的说话人身份验证系统。系统采用了矢量量化识别算法;在码本设计中采用了LBG改进算法;在说话人确认阈值的选取环节中采用了基于统计的先验阈值估计算法和在线递增的阈值更新方法。试验表明,在1.1s内,系统识别率达到92.0%。该系统可嵌入现代建筑或家居的电子类产品中,对实现智能化操作和控制具有较强的实用价值。In order to put the speaker recognition technology into use, an identity recognition system based on DSP was constructed with close small words and text-dependent. This system adopted Vector Quantization, and the improved LBG algorithm was used in the code design. A priori threshold estimation algorithm based on statistics and an on-line in- cremental threshold update method were used in the selection of speaker verification threshold. The experiment results showed that the system could reach a recognition rate of 92.0% within 1.1 seconds. Intelligent operation and control could be well achieved when the recognition system was embedded in electronic products of modem building or house, and it was of strong practical value.

关 键 词：说话人识别 特征参数 数字信号处理 语音身份验证 

分 类 号：TN912.34[电子电信—通信与信息系统]