摘要
对面向"互联网+"的声纹识别技术进行研究,并分析了其在刑事案件侦破中的应用。基于GMM-UBM声纹识别确认系统,对GMM-UBM模型构建方法进行详细描述,研究模型参数最大后验概率算法、估计期望最大化算法、参数训练和识别过程。对基于GMM-UBM的声纹识别系统进行设计,利用对比实验的方法分别验证在相同条件下GMM建模方法、GMM-UBM建模方法的识别效果。在测试随机抽取的一组语音时,系统均具有较高的识别成功率,在进行不同人数测试时,随着样本人数的增加,系统识别率会有少许降低,但平均识别率较高,为89.6%;与GMM系统相比,GMM-UBM系统具有较高的识别率,随着混合度的增加,GMM-UBM系统识别率随之增大。
The voiceprint recognition technology applied to "Internet + " is studied and its application in criminal case investigation is analyzed. On the basis of the GMM-UBM(Gaussian mixture model-universal background model) voiceprint recognition system,the construction method of GMM-UBM is elaborated,and the maximum posterior probability algorithm,estimation expectation maximization algorithm and training & recognition process of model parameters are studied. The voiceprint recognition system based on GMM-UBM is designed. The recognition effects of GMM modeling method and GMM-UBM modeling method under the same conditions are verified respectively by comparative experiments. When testing a group of randomlyextracted voices,both the systems are of high recognition rate. When testing different numbers of people,the recognition rate of the system decreases somewhat with the increase of the number of sampling people,but the average recognition rate is 89.6%,which is high. In comparison with the GMM system,the GMM-UBM system is of higher recognition rate. The recognition rate of GMM-UBM system increases with the increase of the degree of mixing.
作者
魏莲芳
WEI Lianfang(Sichuan Police College,Luzhou 646000,China)
出处
《现代电子技术》
北大核心
2020年第7期34-38,共5页
Modern Electronics Technique
关键词
刑事案件侦破
声纹识别
互联网+
GMM-UBM
识别率
参数训练
criminal case investigation
voiceprint recognition
Internet+
GMM-UBM
recognition rate
parameter training
作者简介
魏莲芳(1975—),女,四川仁寿人,硕士,副教授,研究方向为社会安全与防范技术。
