摘要
为更系统地讨论说话人辨认系统中UBM(universal background model)训练时长对系统识别性能的影响,针对UBM训练时长和混合度设置了一组实验,在基于GMM-UBM(gaussian mixture model-universal background model)的说话人辨认系统中,探讨了UBM训练时长和混合度之间的关系,得出了UBM平均每个混合得到100帧左右训练样本时,能够获得较高且较稳定识别率的结论,并总结出了在某一混合度下UBM训练数据净时长的范围,为以后的研究提供了一个基本的数据依据。
In order to discuss the training duration of UBM’s effect on the properties of recognition system,focusing on UBM training duration and mixture,a series of experiments are made to discuss the relationship between UBM training duration and mixture in speaker identification system based on GMM-UBM（gaussian mixture model-universal background model）.It is concluded that each mixture gaining about 100 frames training sample can get a higher and more stable recognition rate.The range of the time duration of UBM training data in a certain mixture is also summarized.The conclusion provides a basic data basis for future research.
出处
《北京信息科技大学学报（自然科学版）》
2013年第3期87-91,共5页
Journal of Beijing Information Science and Technology University
基金
北京信息科技大学网络文化与数字传播北京市重点实验室开放课题资助项目
