基于统计方法的汉语连续语音中声调模式的研究

A Stochastically-Based Study on Chinese Tone Patterns in Continuous Speech

提出采用决策树的数据驱动方法,结合专家知识,从大规模语料中统计学习出连续语音中声调模式的分布.在建立决策树的过程中,除了相邻音节的声调外,还考虑了多种可能影响声调模式的因素,如音节声韵母发音特点的分类、音节在词中的位置等.决策树建立后,共得到28种声调模式.通过对结果的分析发现,除了上下文的声调外,其它因素对连续语音中声调模式的变化也有一定的影响.声调识别实验的结果证明了该方法的有效性.

A decision tree based approach is proposed for obtaining the quantitative result of tone variation patterns in continuous Chinese speech. While constructing the decision tree, besides neighboring tone, many other possible factors are considered such as syllable position in the word and consonant/vowel type of the syllable, which are not studied in conventional analysis. After the tree is established, 28 different tone patterns and their corresponding model parameters are acquired. From the result it is found that many factors in addition to the tone of neighboring syllable affect tone pattern variation in continuous Chinese speech. Tone recognition experiments demonstrate the effectiveness of this approach.