论文部分内容阅读
为进一步提高文语转换系统中韵律结构预测的准确度,提出了一个基于概率频度的统计模型的方法,预测韵律词和韵律短语边界两级韵律结构。该方法提取与韵律词和韵律短语边界有关的语言学特征(词性、语法词、长度和位置等),并进行样本训练计算各个特征的概率频度值,最终分别建立韵律词和韵律短语的统计模型。实验结果表明:统计模型的方法对于韵律词和韵律短语边界预测的正确率分别可达90.6%和84.6%,并与决策树算法和T ransform ation-based learn ing(TBL)转换规则学习算法比较,提高10%以上的正确率。
In order to further improve the accuracy of prosodic structure prediction in the text-to-speech conversion system, a statistical model based on probability frequency is proposed to predict prosodic words and prosodic phrases. The method extracts the linguistic features (part of speech, grammatical word, length and position, etc.) related to the boundaries of prosodic words and prosodic phrases and conducts sample training to calculate the probability frequency values of each characteristic, and finally establishes the statistics of prosodic words and prosodic phrases respectively model. The experimental results show that the accuracy of the statistical model is 90.6% and 84.6%, respectively, for the prediction of the boundary of prosodic words and prosodic phrases, and compared with the decision tree algorithm and TBL conversion rule learning algorithm, Improve the accuracy of more than 10%.