Recent advances in automatic speech recognition (ASR) technology have enabled us to develop the first commercially available voice name dialing service that provides the user with voice control of all aspects of service operation, including training the ASR device, administering the dialing list, and placing calls to listed parties by saying their name. One of the principal challenges in designing the service has involved anticipating the sources of breakdowns, occasions in which the user is compelled to attend to an ASR capability that has failed to respond as expected. Iterative design-and-test studies have produced a service judged fun and easy to use in recent commercial trials. This paper describes the studies and the design details that have enabled us to avoid or accommodate a variety of potential breakdowns, and so realize the new potential of ASR in telecommunication applications such as voice name dialing.
