AUTOMATING SERVICES WITH SPEECH RECOGNITION OVER THE PUBLIC SWITCHED TELEPHONE NETWORK - HUMAN-FACTORS CONSIDERATIONS

Using speech recognition to automate services over the telephone network is extremely difficult and faces many daunting challenges. There are problems inherent in current telephone transmission and in speaker-independent recognizers, but in this paper we focus on human conversational behavior and human factors issues. Problems encountered by current systems include users' tendencies to speak continuously, rather than pausing between words, to respond with words not in the recognizer's vocabulary, and to speak when the recognizer is not prepared to accept input. Principles of user-interface design are discussed, and a five-stage model for developing new automated systems is presented. The requirements of future systems, which will permit more natural communication between humans and machines, are also discussed, and data on the vocabulary requirements of human conversations are presented.