用语音动态谱特征实现连续汉语语音按音节的自动分割

Realization of Automatic Isolation of Syllables in Continuous Chinese Speech by Using Its Dynamic Spectral Features

本文通过对汉语语音的特性分析,及各类音素的DFT谱特性,特别是清/浊音的DFT谱差异的研究,概括出了可用于连续语音音节分割的两个相对最佳的动态特征;同时,提出了动态特征曲线极小值区域分布情况的一种定量描述方法——凹谷函数描述法。在这些研究的基础上,本文给出了一个具体的分段算法。实验验证表明,本文的分段方法对连续汉语语音的音节分割是有效的。最后,本文将这种方法应用到语图分析中,并首次实现了连续语音动态语图按音节的自动分割。

In this paper two relatively optimum features that can be used in isolating syllables in continuous Chinese speech have been generalized, through characteristics analysis of Chinese speech and the DFT spectral characteristics of various phonemes, especially the DFT distinction of voiced/unvoiced sounds. Meantime it is proposed that the distribution of minima regions of dynamic features curves can be described quantitatively by valley function. Based on these studies, a practical algorithm in segmentation is given. Experimental verification shows that the method of segmentation is effective in isolating the syllables in continuous Chinese speech. Finally this method is applied to the analysis of sonogram furthermore the automati cisolation of syllables in the dynamic sonogram of continuous speech is first realized.