ISSN:
1572-8110
Keywords:
prosody prediction
;
parsing
;
tone groups
;
superposition of pitch patterns
;
TTS synthesis
Source:
Springer Online Journal Archives 1860-2000
Topics:
Linguistics and Literary Studies
,
Computer Science
Notes:
Abstract Proper prosodic structure is crucial for natural-sounding synthesized speech. Because of the lack of other information on discourse structure, we have to rely on syntactic structure in order to predict the main prosodic items for normal speech. To meet this requirement, a dependency-based parser has been developed for Hungarian that assigns the boundaries of functional constituents in the sentence, in other words, the places where new intonation patterns start and breaks can be inserted. We determine stress distribution in the sentence, using four levels including focus. The practical realization of the prosodic predictor also relies on statistical and empirical data. The intonation units (tone groups) with proper melody (e.g., falling, slowly falling, level, rising, slowly rising, rising-falling, and falling-rising) are established on the base of syntactic properties in declarative, interrogative, and imperative sentences. The results are embedded in an experimental Hungarian text-to-speech (TTS) system.
Type of Medium:
Electronic Resource
URL:
http://dx.doi.org/10.1023/A:1026519300902
Permalink
