端点检测的不准确是造成语音识别错误的一个重要原因,根据元音和噪音信号频能比的差异.以及辅音信号的过零率特点,论文提出了一种改进的基于频能比的端点检测算法——MEPD—FER.实验表明该算法可以在有噪音存在的情况下快速准确地确定出语音信号的端点。
According to the difference between the FER of vowels and noises,and the characteristics of the ZCR of consonants,it is proposed in this paper that a modified algorithm endpoint detecting of speech using FER.And it is demonstrated that this algorithm is effective to detect the endpoint of speech in the presence of noise.