An automated method for collecting speech samples from subjects speaking over ordinary dialed-up telephone lines is described. Callers enter into a dialogue with an applications machine which is a computer system with both word recognition and speech output capabilities. The first part of the interaction involves a semi-formal training procedure during which the callers are asked to repeat a set of words in order to produce templates. In the second part of the session, the machine asks the callers certain questions, and the subjects' responses are recognised and recorded. Thus examples of words spoken in the context of a true machine-human dialogue are obtained.
展开▼