Automatic accentuation and prosodic phrasing for dutch text-to-speech conversion

Correct accentuation and phrasing improves the quality of synthetic speech. This paper discusses an algorithm which assigns both sentence accents and phrase boundaries on the basis of the prosodic sentence structure. Although this latter structure is theoretically derived from the syntactic structure, the present algorithm deterrnines the prosodic structure by means of (linguistically and statistically motivated) rules of thumb, i.e., without exhaustive syntactic analysis. Subsequently, both prase boundaries and sentence accents are assigned on the basis of this prosodic sentence structure. In this component, several factors affecting accentuation are imitated by rules which do not directly refer to the syntactic or thematic structure.