The MaryTTS entry for the Blizzard Challenge 2016

Sébastien Le Maguer, Ingmar Steiner

In: Blizzard Challenge. SynSIG Blizzard Challenge (SSW) located at ISCA Tutorial and Research Workshop on Speech Synthesis September 16-16 Cupertino CA United States 2016.


The MaryTTS system is a modular architecture text-to-speech (TTS) system whose development started around 15 years ago. This paper presents the MaryTTS entry for the Blizzard Challenge 2016. For this entry, we used the default configuration of MaryTTS based on the unit selection paradigm. However, the architecture is currently undergoing a massive refactoring process in order to provide a more fully modular system. This will allow researchers to focus only on some part of the synthesis process. The current participation objective includes assessing the current baseline quality in order to evaluate any future improvements. These can be achieved more easily thanks to a more flexible and robust architecture. The results obtained in this challenge prove that our system is not obsolete, but improvements need to be made to maintain it in the state of the art in the future.


Weitere Links

MARYTTS_Blizzard2016.pdf (pdf, 238 KB )

German Research Center for Artificial Intelligence
Deutsches Forschungszentrum für Künstliche Intelligenz