语音分组识别技术的研究  

Research on Speech Recognition by Group Technology

作　　者：李云[1] 鲍鸿[1] 

机构地区：[1]广东工业大学自动化学院,广东广州510006

出　　处：《广东工业大学学报》2014年第2期54-57,共4页Journal of Guangdong University of Technology

基　　金：教育部青年基金资助项目(10TJCZH220)

摘　　要：为了减少语音识别时间,降低系统资源耗费,提出一种针对非特定人、孤立词、大词汇量的语音分组识别算法.运用K均值聚类算法对语音分组,并对语音分组特征进行置信度检验,使分组稳定,保证分组后识别率不下降.通过对非特定人孤立词的语音识别的实验,证实了该方法的有效性.In order to reduce the time of speech recognition and the consumption of system resources,it proposed a method of speech recognition by grouping according to the speaker-independent,isolated words and large vocabularies.The method was based on K-means clustering,and the characteristics of the group were inspected by the confidence in order to make the group stable and ensure that the rate of recognition does not fall after grouping.The experiment results verify the effectiveness of this method.

关 键 词：Mel频率倒谱特征参数 K均值聚类 置信度 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]