Machine Learning algorithms for linguistic aspects of speech synthesis

The goal of this project is the data-oriented construction of a computer model of the linguistic knowledge and processes relevant for text to speech conversion. Especially the determination of lexical stress, prosodic phrasing of sentences, and the determination of sentence accents are studied. The method we propose uses inductive artificial learning algorithms. A second goal of this project is to acquire a better insight into how and in what proportion information sources, which are claimed to be necessary in the literature, contribute effectively to a naturally sounding intonation, thereby gaining linguistic relevancy.