摘要
以Xilinx公司Virtex-IIPro为开发平台,实现了一个基于隐马尔可夫模型(Hidden Markov Model,简称HMM)非特定人的孤立词语音识别系统。系统采用改进的基于语音对数域能量变化率的实时端点检测算法,仅对检测的有声段语音进行特征提取和解码,减少了要处理的语音帧数。实验表明系统在150词条的情况下识别率达到97.3,识别时间为1.42倍实时。
An embedded speaker-independent isolated word speech recognition system is designed and realized in the Xilinx Virtex-II Pro platform. With the help of a modified real time voice activity detection algorithm （VAD） based on the log. Energy acceleration associated with voice onset, feature extraction and decoding to the active voice and ignore the frames of non-activity are performed. Test on150 words vocabulary shows that system provides a recognition accuracy rate of 97.3 %using only 1.42 times of real time.
出处
《信息技术》
2008年第12期89-92,共4页
Information Technology
关键词
隐马尔可夫模型
语音识别
端点检测
语音系统
FPGA
hidden markov model
speaker recognition
speech endpoint detect
speech recognition system
FPGA
