Making a Czech synthesizer talk German

Eva Lasarcyk

The motivation of the project was to find out, how well a corpus-based speech engine used for the synthesis of Czech would manage to synthesize a foreign language, given a speech database of that language.

During a visit at the Laboratory of Speech and Dialogue in Brno (Czech Republic), a speech corpus of single German words was designed and recorded. The words were then segmented into phones so they could serve as a database for the (originally Czech) unit selection speech synthesizer.

The talk will comprise some general information about speech corpora, speech synthesis, and the principles of the speech engine used. The different steps of the database construction will be described, as well as changes in the strategy of segmentation during the working process. Several sound examples used for evaluation purposes will be presented to demonstrate the difficulty of making a Czech synthesizer talk German.