Sixth European Conference on Speech Communication and Technology
Naturally sounding synthesized speech requires proper prosodic structure. The unequivocal relation between syntax and prosody is contestable, but for lack of other information on discourse structure, we have to rely on syntactic structure in order to determine some prosodic features. This work - based on basic research results in Hungarian linguistics - started with a preliminary parser for simple Hungarian sentences. It tries to establish phrase boundaries with possible breaks, and to determine the focus as well as other stressed and unstressed elements The current goal is to automatically predict the above prosodic features for complex sentences. The steps of the prosody module are: pre-processing, morphological analysis, phrase parsing, sentence parsing, search for the focus, stress and break distribution. The prosodic predictor is implemented in the MULTIVOX multilingual text-to-speech system.
Full Paper (PDF)
Bibliographic reference. Koutny, Ilona (1999): "Parsing hungarian sentences in order to determine their prosodic structures in a multilingual TTS system", In EUROSPEECH'99, 2091-2094.