Abstract: This paper describes the two Text-to-Speech (TTS) systems presented by Aholab-EHU/UPV in the Albayzin2010 evaluation campaign. The first system is a statistical parametric TTS based on HTS, with the incentive of using our own vocoder. The other one is a hybrid system in which we try to take advantage of the consistency of the statistical averaging and the segmental naturalness of the unit selection approach. It uses the acoustic parameters generated by the statistical system as the target sequence during the unit selection process. Informal listening tests and some objective measures show that adding the Intonation Break information during the voice building process improves the performance of both systems.
Index Terms: speech synthesis, statistical parametric, unit selection, evaluation.