A Minimum‐Distance Search Technique and its Application to Automatic Directory Assistance

This paper describes a new search procedure and its application to the problem of obtaining telephone directory information from spoken spelled input. The method obtains its speed from using the concept of equivalence classes, with names classified according to their letter‐by‐letter acoustic similarity. It derives its accuracy from the use of a minimum‐distance criterion for selecting answers. The search procedure finds the name with the minimum distance, usually after only a small fraction of the directory file has been examined. Using an acoustic analyzer with an 80 percent correct recognition rate for individual letters, a 98.6 percent correct recognition rate for names was achieved when the method was applied to a directory of 18,000 entries. On the average, only 1.2 percent of the directory had to be examined for each query. With an input recognition rate of 71 percent for letters, the respective figures were 97.2 percent and 2.8 percent.