摘要： 相关向量机(RVM)分类法使用概率输出克服了支持向量机(SVM)识别速率低的缺点，并且具有更好的稀疏性。但在与文本无关的话者辨别中，大量训练样本数据体现了RVM在模型训练时计算量与内存需求过大的缺点。针对以上特点，提出基于GMM统计特征参数与RVM融合的与文本无关的语者辨别系统，既有效地提取话者特征信息，解决大样本数据下的RVM训练问题，又结合统计模型鲁棒性高和分辨模型辨别效果好的优点。实验结果证明，该系统比基本的GMM系统具有更优的错误辨别率，比GMM/SVM系统具有更高的稀疏性。

关键词: 相关向量机, 高斯混合模型, 话者辨别, 支持向量机

Abstract: Relevance Vector Machine(RVM) classification method uses the probabilistic output to overcome Support Vector Machine(SVM) shortage as well as has more sparsity. Whereas RVM has overloaded computation complexity and memory storage when it is applied for the text-independent speaker identification because of the mass training samples. For solving the problem, a hybrid GMM and RVM approach is proposed which effectively abstracts the speaker feature vector as well as solve the mass storage problem. Further more, this hybrid approach combines the robustness of generative model and the powerful classification of discriminative model to improve the performance and robustness of identification. Experimental results prove that the method has better error identification rate than the GMM system and more sparsity than state-of- the-art GMM/SVM system.

Key words: Relevance Vector Machine(RVM), Gaussian Mixture Model(GMM), speaker identification, Support Vector Machine(SVM)

中图分类号: 

• TP18