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.
出处
《科技通报》
北大核心
2012年第4期57-59,共3页
Bulletin of Science and Technology
作者简介
熊于菽（1979-），女，重庆沙坪坝人，硕士，讲师，研究方向：语音识别、信号与信息处理，图像处理。
