基于GMM统计特性参数和SVM的话者确认

Text-Independent Speaker Verification Based on GMM Statistical Parameters and SVM

针对与文本无关的话者确认中大量训练样本数据的情况 ,本文提出了一种基于 GMM统计特性参数和支持向量机的与文本无关的话者确认系统 ,以说话人的 GMM统计特性参数作为特征参数训练建立目标话者的SVM模型 ,既有效地提取了话者特征信息 ,解决了大样本数据下的 SVM训练问题 ,又结合了统计模型鲁棒性好和辨别模型分辨力好的优点 ,提高了确认系统的确认性能及鲁棒性。对微软麦克风语音数据库和 NIST′0 1手机电话语音数据库的实验表明该方法的有效性。

A system based on Gaussian mixture model (GMM) statistical parameters and support vector machine (SVM) for text independent speaker verification is proposed aimed at a large quantities of data. The SVM speaker model is trained from the parameters of GMM, which extracts the speaker feature effectively and does well in a large quantities of data. Moreover, the system combines the robustness of generative model and the powerful classification of discriminative model to improve the performance and robustness of verification. Text independent speaker verification experiments are conducted on the telephony 2001 NIST Speaker Recognition Evaluation corpus and handset microphone corpus provided by Microsoft Research Asia. The system is validated to be effective.