基于虚拟仪器技术的VQ-SVM并行说话人识别系统  

The VQ-SVM Parallel Speaker Recognition System Based on the Virtual Instrument Technique

作　　者：刘祥楼[1,2] 吴香艳[1,2] 高丙坤[1] 

机构地区：[1]东北石油大学 [2]黑龙江省高校校企共建测试计量技术及仪器仪表工程研发中心,大庆163318

出　　处：《科学技术与工程》2012年第11期2590-2593,共4页Science Technology and Engineering

基　　金：黑龙江省教育厅科学技术研究项目(11551002)资助

摘　　要：说话人识别混合方法是目前研究的热点。基于虚拟仪器技术并融合说话人识别技术,提出矢量量化和支持向量机方法结合,依托MATLAB实现运算,由LabVIEW以多任务管理和调用MATLAB来实现说话人并行识别处理。经自建小样本语料库仿真实验,结果表明:系统识别率98.54%、误识率5.28%、识别时间0.25 s,较单一矢量量化和支持向量机方法识别率分别提高了3.66%和1.16%,误识率分别降低了6.01%和4.43%。随着样本数的增多,矢量量化方法识别率呈上升趋势,而支持向量机方法识别率呈下降趋势。由此可见:两种方法优势互补实现并行识别可提高系统主体性能。Speaker recognition mixed methods are a hot research at present.Based on virtual instrument technique and integration of speech recognition technique,advanced the method of mixed the Vector Quantization and the Support Vector Machine,the arithmetic supported by MATLAB,the LabVIEW multi-tasks managed and called MATLAB to achieve the speaker＇s parallel recognition processing.Using the private small-scal corpus speaker recognition simulation and test,the results indicated that： the recognition ratio of the parallel system is 98.54%,the recognition error is 5.28%,the testing time is 0.25 s,comparative the single VQ method and single SVM method the recognition ratio is separatly increased 3.66% and 1.16%,the recognition error is separatly reduced 6.01% and 4.43%;with the increase of the numbers of samples,the recognition ratio of VQ method is upward trend,while recognition ratio of the SVM method is downward trend.It is now clear that as two methods having complementary advantages,parallel recognition and can improve the systematic major capabilities.

关 键 词：说话人识别 矢量量化 支持向量机 并行识别系统 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]