一种改进的二阶M-KWS关键词识别方法  

An Improved Second Order Keyword Spotting Model

作　　者：熊于菽[1,2] 冉晟伊[1] 冯晓荣[2] 

机构地区：[1]重庆工程职业技术学院 [2]重庆大学通信工程学院,重庆400037

出　　处：《科技通报》2012年第4期57-59,共3页Bulletin of Science and Technology

摘　　要：在关键词检出系统中,关键词权值的不准确性对系统识别率有很大影响。为了避免关键词权值估计错误,提出了基于上下文相关最大后验概率准则(CDMAP)的改进二阶M-KWS模型,有效地解决了关键词权值的正确估计问题。采用消除内插算法与关键词确认模型相结合,解决了参数估计中的"零概率"问题,提高了系统的检出率。实验表明,该模型在保证误警率较小的情况下,系统关键词识别效果得到有效改善。Keyword spotting system detection performance degrades due to keyword weighting inaccuracy. An improved second order M-KWS method based on CDMAP was proposed. The combination of deleted interpolation algorithm with keyword recognition model can efficiently solve the ＂zero probability＂ problem during the parameter evaluation. Experiments showed that the detection performance of the improved system is greatly enhanced.

关 键 词：上下文相关最大后验概率 关键词权值 消除内插 关键词检出 

分 类 号：TN912.34[电子电信—通信与信息系统]