Publication at Faculty of Mathematics and Physics |
2006
Abstract
We have recently introduced our new data-driven prosody model for text-to-speech synthesis. The model is based on text parametrisation using a generative prosodic grammar and on automatic speech corpora analysis methods.
The model has been successfully tested for Czech language with highly assessed naturalness of resulting speech. The current stage of the model development involves the first experiments with its application to German synthesis.
The results of these experiments are presented in the paper.