The Recording of Emotional Speech - JST / CREST database research - Nick

This paper reports on a recent development in Japan, funded by the Science and Technology Agency, for the creation and analysis of a very large database, including emotional speech, for the purpose of speech technology research. In order to capture spontaneous samples of naturally-occurring emotion without recourse to acting, a very large amount of conversational speech and daily spoken interaction will be recorded and labeled. The algorithms developed for automating the labeling of this speech will form the foundation for user-interactive speech processing technology that is sensitive to speaking style and speaker-state. This paper addresses the issue of collecting such a large amount of natural speech data.