摘要
针对多说话人聚类线性初始化方法精度较差的问题,提出了一种改进的聚类初始化方法。该方法引入BIC对由线性初始化产生的初始类进行检测分割,有效提升了说话人初始类纯度。最后将该方法应用到高斯混合模型(GMM)多说话人识别系统。实验结果表明,所提方法使说话人平均类纯度(ACP)提高了48.51%,系统的错误识别率平均降低12.09%。
Aiming at the problem of the linear initialization method of multiple speaker clustering with poor accuracy,this paper proposed an improved method of clustering initialization.The method by introducing BIC to detect and segment for initial cluster produced by the linear initialization,and promoted the purity of speaker initial cluster effectively.Finally,applied the method to Gaussian mixture model(GMM) multi-speaker recognition system.And the experimental results show that this proposed method makes the average cluster purity(ACP) have been increased by 48.51%,and the error recognition of system have been reduced by 12.09% on average.
出处
《计算机应用研究》
CSCD
北大核心
2012年第2期590-593,共4页
Application Research of Computers
基金
甘肃省财政厅资助项目(0914ZTB148)
甘肃省自然科学基金资助项目(1014ZSB064)
关键词
多说话人识别
改进的聚类初始化
高斯混合模型
平均类纯度
multi-speaker recognition
improved clustering initialization
Gaussian mixture model
average cluster purity