智能上假肢特定人语音识别系统实现

摘要

The speaker-independent speech recognition system for intelligent prosthetics arm adopted speaker-independent speech recognition technology to enable real-time and accurate control of prosthetics arm.Real-time display of speech waveform was implemented.Uniform sub band variance and double threshold were used to implement endpoint detection in noisy environments.Mel filter bank,I-Mel filter-banks and Mid-Mel filter-banks were used in the same time for avoiding missed information.Using global minimum of DTW method recognized reference templates and speech commands.Results of test show that the short-term average recognition rate is 97.875 ％ and the phrase average recognition rate is 97 ％.Thus,the system achieves real-time control and accurate identification,meeting application needs.%智能上假肢特定人语音识别系统采用特定人语音识别技术对智能上假肢进行实时准确控制.实现对语音波形的实时显示;运用均匀子带方差和双门限方法,实现噪音环境下特定人语音的实时端点检测;引入I Mel滤波器和Mid-Mel滤波器对传统Mel滤波器进行改进,有效提高信息完整度;采用全局最小值DTW算法对参考模板和待识别指令进行匹配识别.测试结果表明,该系统对短词的平均识别率为97.875％o,短句的平均识别率为97％.实现了对控制指令实时、准确的识别,满足系统应用需要.