Arquitectura para conversión texto-habla multidominio

this paper presents an evolution in the design of multi-domain unit selection text-to-speech (MD-TTS) architecture. The goal of this approach is to achieve good synthetic quality like the limited domain TTS systems, without discarding a general purpose synthesis. The multi-domain architecture entails a speech corpus containing several domains. Moreover, it has been hierarchically structured by means of independent component analysis in order to optimize the unit selection process. On the other hand, MD-TTS demands a module capable to classify the texts in multiple domains, considering the system requirements.