Speech Recognition Vocabulary Scoping for Automated Call Routing

Abstract

Call routing involves directing incoming telephone calls from a central number to an appropriate person or department. In the course of an ongoing work project, a quick study was performed to scope the vocabulary requirements for a speech recognition automated call routing application for a large department store. Forty-one participants were given 35 sample shopping tasks and were asked which department they would ask for when calling the store. The range of responses for a given task was large. With a 29 item recognition vocabulary consisting of most frequent responses and root phrases (e.g., “sport” for “sporting goods”), 57% of user responses would be covered. Users were also asked to rate the confidence of their department choice. The greater the variety of responses to a task across all participants, the less confident participants were of their responses.