Designing a speech corpus for instance-based spoken language generation

Abstract

In spoken language applications such as conversation systems where not only the speech waveforms but also the content of the speech (the text) need to be generated automatically, a Concept-to-Speech (CTS) system is needed. In this paper, we address several issues on designing a speech corpus to facilitate an instance-based integrated CTS framework. Both the instance-based CTS generation approach and the corpus design process have not been addressed systematically in previous researches.