Combination of text level defined prosody elements with word accents to realise an improved German prosody for text-to-speech

摘要

This work was completed in the framework of the SPEAK! project (Copernicus No. 10393). The aim of this development was to improve the quality of synthesised speech output to be used in dialogue systems as an additional element of multimodal machine-man interfaces. A special text generator component works in the dialogue system to predict the basic intonation contour types, phrase boundaries, phrase intonation contours types and focus words in phrases. A marker set was defined to embed this information into the plain text to be synthesised. The expectation of this so uation is to improve sentence intonation and functionality of a German text-to-speech system. This text level marker set (4 basic parameters) serve - as a carrier skeleton - for the realisation of the final prosody structure in the sentence. The phonetic level prosody module combines these 4 text level parameters with the word accent and prosody generation process during the final speech generation. A new generation of internal rule set (both for segmental and suprasegmental processing) was designed and realised for the MULTIVOX text-to-speech system to generate high quality German synthetic speech. In this report only the latter will be discussed.