Recording techniques for capturing natural every-day speech

This paper describes techniques for the collection of natural spontaneous speech from daily conversational interactions for a large corpus that is currently being produced by the Japan Science and Technology Agency. This corpus will form the basis for further development of tools and software for the improvement of concatenative speech synthesis and for the development of spoken-language interfaces for information-providing devices that will be sensitive not only to the content of an utterance, but also to the manner in which it is spoken, so as to be able to detect speaker emotions and attitudes.