Second ESCA/IEEE Workshop on Speech Synthesis
September 12-15, 1994
This paper describes a new prosodic phrasing implemented on the German version of the multilingual CNET PS OLA TTS system. The new approach renders parsing reliable and robust. Moreover it provides efficient results for the majority of text in the context of unlimited vocabulary and unrestricted syntactical complexity, where neither semantic nor pragmatic information is available.
The whole process is split into two steps, firstly, the decomposition of the sentences (identified by punctuation marks) into minimal syntactic units; and secondly, the arrangement of the syntactic units in prosodic phrases (constituents).
The use of minimal syntactic units avoids the obligatory construction of a syntax tree, which is a well-known source of errors. Since stress rules usually start from the syntax tree, this approach demands an alternative to supplying degrees of accentuation. In this approach, each rule for the parsing into minimal syntactic units implicitly denotes those degrees.
Bibliographic reference. Frenkenberger, S. / Schnabel, Betina / Alissali, M. / Kommenda, Markus (1994): "Prosodic parsing based on parsing of minimal syntactic structures", In SSW2-1994, 143-146.