摘要
在语音识别系统中产生错误识别的原因之一是端点检测有误差。针对短时过零率对噪声的存在非常敏感,本文引入一种判决门限,修正了传统过零率的计算。同时引入窗长动态改变的端点检测方法,并将两者有机的融合到传统的双门限端点检测算法中。试验表明这种算法可以比较精确的检测出语音端点,适合于对端点检测比较敏感的语音识别算法。使用改进后的语音端点检测方法,可以有效地提高语音识别率。
A major cause of errors in speech recognition systems is the inaccurate endpoint detection. Aim at the sensitivity of exist- ing of noises, we amend the traditional Zero-crossing measurements by introduce decision thresholds. Incorpor ated the algorithm with dynamic length of windows, the traditional algorithm with double restricted values modified by introduce decision threshold are applied to endpoint detection of speech signals. Experimental results show the efficiency of the new algorithm, especially on algorithms sensitive to endpoint detection.
出处
《微计算机信息》
北大核心
2006年第05S期138-139,共2页
Control & Automation
基金
四川省科技攻关项目.编号:04GG006-025
关键词
语音识别
语音信号处理
端点检测
Speech recognition
Speech signal processing
Endpoint detection.
