论文部分内容阅读
焦点是语言表达的重要方式,焦点重音是重要的韵律特征,实现中性语音到焦点语音的转换可以提高语音的表现力。该文提出了声学特征局部凸显度的表示方法,分析了由中性语音到焦点语音,焦点单词所属音节声学特征变化与中性语音相应音节声学特征局部凸显度的相关性,提出了一种基于决策树的英语焦点语音的转换模型。该模型采用决策树对训练语料进行聚类,所用上下文包括音节与焦点单词的相对位置以及音节在韵律结构(如韵律短语、韵律词等)中的位置。在此基础上,提出了一种基于局部凸显度的中性语音到焦点语音声学特征变化的预测算法。采用该算法后,客观实验中声学特征变化平均绝对值误差降低到0.08,主观实验表明本文提出的模型的转换语音具有更好的焦点表达效果和自然度。
Focus is an important way of language expression. Focus stress is an important prosodic feature. The conversion from neutral voice to focus voice can improve the voice expressiveness. In this paper, we present a method to represent the local prominence of acoustic features. Based on the analysis of the correlation between the acoustic features of syllables from the neutral speech to the focus speech and the local convexity of the acoustic features of the corresponding syllables of neutral speech, The Conversion Model of English Focus Voice in Decision Tree. The model uses the decision tree to cluster the training corpus. The context used includes the relative positions of syllables and focus words and the positions of syllables in prosodic structures (such as prosodic phrases, prosodic words, etc.). Based on this, a prediction algorithm based on local prominence for the change of neutral speech-to-focus phonetic features is proposed. With this algorithm, the average absolute error of the acoustic features in the objective experiment is reduced to 0.08. The subjective experiments show that the speech of the proposed model has better focus expression and naturalness.