嵌入式数字语音拨号系统的研究与实现  

Research and Implementation of Speaker-Independent Digital Voice Dialing System

作　　者：李芬兰[1] 马小月[1] 

机构地区：[1]汕头大学电子信息工程系,广东汕头515063

出　　处：《电声技术》2012年第1期46-50,共5页Audio Engineering

摘　　要：将语音识别技术应用于拨号系统,在嵌入式平台上实现了一款针对非特定人的数字语音拨号系统。语音识别算法中选择梅尔频率倒谱系数为特征参数,连续隐马尔科夫模型。为训练和识别过程模型,利用Qt界面对识别过程进行控制,系统针对非特定人数字语音识别进行实验。结果表明,系统针对非特定人识别率达到了98%,识别时间为3.55 s。识别率和实时性都满足语音拨号的需求。Speech recognition technology is put into practice, a speaker- independent digital speech dial system is intro- duced. In the speech recognition process, the MFCC is used as the feature parameter and the CHMM is chosen as the train- ing and recognition model. The recognition by Qt GUI is controled and this system on speaker - independent digital recogni- tion is tested. The result indicates the accuracy of this system reaches up to 98% to speaker - independent ease and it costs 3.55s. It can meet both the accuracy and time requirements of speech recognition system.

关 键 词：嵌入式语音拨号系统 连续隐马尔科夫模型 梅尔频率倒谱系数 Qt界面 

分 类 号：TN912.3[电子电信—通信与信息系统]