中文摘要：

为避免“绝对”声韵分割策略的主观性和随意性，结合语谱图以及匹配追踪算法，实现了一种对汉语孤立字进行重叠声韵分割的新的时频方法。以语谱图判决得到的浊音起点为声韵母过渡段的起点，以匹配追踪原子参数在浊音起点之后所达到的第一个极值的位置为过渡段终点。仿真实验结果表明，该方法的分割正确率可达87．5％；将分割后的声韵母单元分别送入语音识别系统，与以整个字节为识别单元相比识别率提高了1．33％。

英文摘要：

The spectrogram and matching pursuit （MP） algorithm are combined to avoid the subjectivity and randomness caused by ＂absolute＂ segmenting scheme for Chinese syllable, which resulted in a new overlapped segmenting method. The outset of the transition section is determined by the vowel outset judged from spectrogram, while the ending is treated as its corresponding MP atom parameters~ first extremum position after the outset above. As is shown in experiments, the segmenting accuracy is 87. 5%; the segmented initials and finals are input the speech recognition system, and achieved an increment by 1.33% is achieved to compared with recognition with syllable units.