基于关联规则的藏语语音韵律参数提取

Extracting Tibetan prosodic patterns by association rule

韵律规则对于语音识别和语音合成具有重要意义,韵律特征参数的描述正确与否直接影响合成系统的输出。为了提高藏语语音合成中语音的自然度,本文研究了基于数据挖掘中的关联规则来发现韵律参数之间的相互关系,并基于关联规则算法获得藏语韵律参数中基频参数的变化规则,这些规则可以为藏语语音合成系统的选音提供帮助。

The pitch models play an important role in speech recognition an synthesis. Prosodic parameters choice right or wrong influences the quality of synthetic speech directly. In this paper, we discover those relationships by association rules modeling and generate association rules of prosodic parameters including pitch parameters. The output rules are able to offer convenience in Tibetan speech system.