This paper describes a speaker-independent word spotting algorithm and its application in a voice activated service name entry system over telephone networks. The system allows a user to specify a keyword with extraneous speech in response to system audio prompts. A task oriented multi-speaker isolated-word database is used to construct keyword HMMs. The score of keywords is computed by using the continuous Viterbi decoding algorithm with score normalization using a background HMM. The candidates for keywords are obtained after evaluating the partial matches. The dialogue design of the system improves user-friendliness and shortens the user's operation time by employing a newly proposed adaptive confirmation procedure. The procedure effectively repeats commands and masks inappropriate control functions. An evaluation over real telephone lines is presented that confirms improved user interface for menu selection tasks.

Keywords: Word spotting, Dialogue system