Applications and Science in Soft Computing

0405193 - UIVT-O 20030201 RIV DE eng M - Část monografie knihy
Šebesta, Václav - Tučková, J.
Optimisation of Neural Network Topology and Input Parameters for Prosody Modelling of Synthetic Speech.
Applications and Science in Soft Computing. Berlin: Springer, 2004 - (Lotfi, A.; Garibaldi, J.), s. 9-16. Advances in Soft Computing, 24. ISBN 978-3-540-40856-7
Grant CEP: GA ČR GA102/02/0124
Výzkumný záměr: AV0Z1030915
Klíčová slova: feature selection * text-to-speech processing * prosody modelling * speech synthesis * neural network utilization
Kód oboru RIV: BB - Aplikovaná statistika, operační výzkum

We try to investigate methods for extracting knowledge from existing continuous speech databases for optimisation of neural network topology to improve the generalization ability of ANN. The ANN for the modelling of fundamental frequency and duration of a speech unit for a Text-to-Speech synthesis are trained by natural speech. The principle of synthesizer is based on the concatenation of speech units. We also use speech unit segmentation of a text for prosody modelling.
Trvalý link: http://hdl.handle.net/11104/0125388