Přejít na obsah

Publikace

Detail publikace

Citace

Tučková, J. and Matoušek, J. : Czech language features selection and prosody modelling for text-to-speech synthesis . ECMS 2003, p. 98-102, Technical University , Liberec, 2003.

Abstrakt

Speech has probabilistic behaviour. It is very difficult to define its specific properties. In that case we can use statistic methods, e.g. Artificial Neural Networks (ANN) and General Unary Hypotheses Automaton (GUHA). In this paper, the replies to two aspects, which affect the speech naturalness, are searched. They are a selection of most important parameters, which play important role for prosody modelling. The second one is an influence of co-articulation for the fundamental frequency (F0) correct values of phonemes determination. The application of ANN for the fundamental frequency and duration (D) of phonemes modelling, the minimization of the number of input parameters, the reduction of the structure redundancy by GUHA and pruning methods and Czech synthesizer ARTIC for the result verification were used.

Detail publikace

Název: Czech language features selection and prosody modelling for text-to-speech synthesis
Autor: Tučková, J. ; Matoušek, J.
Jazyk publikace: anglicky
Datum vydání: 2.6.2003
Rok vydání: 2003
Typ publikace: Stať ve sborníku
Název časopisu / knihy: ECMS 2003
Strana: 98 - 102
ISBN: 80-7083-708-X
Nakladatel: Technical University
Místo vydání: Liberec
Datum: 2.6.2003 - 4.6.2003
/ /

Klíčová slova

text-to-speech, neural network, prosody training, prosody modelling, prosody parameter selection

BibTeX

@INPROCEEDINGS{TuckovaJ_2003_Czechlanguage,
 author = {Tu\v{c}kov\'{a}, J. and Matou\v{s}ek, J.},
 title = {Czech language features selection and prosody modelling for text-to-speech synthesis},
 year = {2003},
 publisher = {Technical University },
 journal = {ECMS 2003},
 address = {Liberec},
 pages = {98-102},
 ISBN = {80-7083-708-X},
 url = {http://www.kky.zcu.cz/en/publications/TuckovaJ_2003_Czechlanguage},
}