摘要
为了解决传统高斯混合模型中期望值EM处理必须具备足够数量的样本才能开始训练的问题,提出了一种新的高斯混合模型在线增量训练算法。本算法在Ueda等人提出的Split-and-Merge EM方法基础上对分裂合并准则的计算进行了改进,能够有效避免陷入局部极值并减少奇异值出现的情况;通过引入时间序列参数提出了增量EM训练方法,能够实现增量式的期望最大化训练,从而能够逐样本在线更新GMM模型参数。对合成数据和实际语音识别应用的实验结果表明,本算法具有较好的运算效率和分类准确性。
This paper presented a new online incremental training algorithm of Gaussian mixture model （GMM） ,which aimed to update GMM model parameters online incrementally instead of waiting for a block of data with the sufficient size to start training as in the traditional EM procedure. The proposed method was extended on split-and-merge EM procedure by Ueda with a new merge and split operation,so inherently it was also capable to escape from local maxima and reduce the chances of singularities. By introducing the time sequence to all the model parameters,also proposed a new online incremental EM training algorithm to update GMM model parameters sample by sample. Experiments on the synthetic data and speech processing task show the advantages and efficiency of the proposed method.
出处
《计算机应用研究》
CSCD
北大核心
2010年第8期2906-2908,共3页
Application Research of Computers
基金
国家"863"计划资助项目(2007AA11Z249)
上海市科委自然科学基金资助项目(08ZR1409300)
上海市重点学科建设项目(S30602)
关键词
高斯混合模型
在线训练
分裂融合算法
模式分类
Gaussian mixture model（GMM）
online training
split-and-merge
pattern classification
