Abstract: Accurate segmentation of the I/F（Initial and Final） is a very important step in Chinese continuous speech recognition based on I/F.Current I/F segmentation methods achieve high degree of accuracy in non-continuous speech，but fall in continuous speech.In this paper，a new method based on entropy and the formant energy of Chinese vowels is proposed and overlapped segmenting scheme is introduced，which can accurately segment the I/F.The results of experiments show that this method achieve high rate of success even in low SNR.

Key words: continuous speech recognition, entropy, formant

摘要： 在基于声韵母的汉语连续语音识别中，声韵母基元的准确分割是非常重要的一步。当前的声韵母分割方法在非连续语音中分割准确度较高，但在连续语音中准确度大幅度降低。利用熵与汉语元音的共振峰能量设计了一种新的分割方法，并引入重叠分割策略，完整地分割出汉语连续语音中的声韵母基元。实验结果表明，这种分割方法在低信噪比下也有很高的分割正确率。

关键词: 连续语音识别, 熵, 共振峰