改进的Apriori算法对语音合成中韵律参数的预测

Predicting Prosodic Parameters for Speech Synthesis with Improved Apriori Algorithm

在语音合成系统中,韵律特征参数的描述正确与否直接影响着语音合成系统的输出效果。为了 解决目前语音合成技术中提高合成语音自然度这一难点,研究了数据挖掘技术来发现韵律参数之间的 相互关系。通过改进Apriori算法对这些关系进行规则描述,来解决目前语音合成中缺陷的前后音节的 韵律参数之间的关系。结果表明,经改进后适合语音数据的Apriori算法,可以获得汉语韵律参数中基频 参数和时长参数的变化规则,这些规则较方便的为语音合成系统的选音提供帮助和指导。

In speech synthesis, Prosodic parameters choice fit influences the quality of synthetic speech di- rectly. Data mining technology is proposed in this paper to define the relationships through association rules modeling in order to overcome the current technical difficulty in improving the spontaneousness of speech sy＇nphesis. We have developed an improvement to apriori algrithm to resolve text to speech system , which has less effective description to reflect data relationships in the corpus. Research findings prove that an improvement to apriori algorithm for generating association rules of prosodic parameters including pitch parameters and duration parameters from corpus and for syllable choice is able to offer convenience in speech system.