On the Recognition of Isolated Digits From a Large Telephone Customer Population

Research on the problems involved in speech recognition has been carried out at Bell Laboratories for close to a decade. 1-7 In all these *Bell Laboratories. °Copy right 1983, American Telephone & Telegraph Company. Photo reproduction for noncommercial use is permitted without payment of royalty provided that each reproduction is done without alteration and that the Journal reference and copyright notice are included on the first page. The title and abstract, but no other portions, of this paper may be copied or distributed royalty free by computer-based and other information-service systems without further permission. Permission to reproduce or republish any other portion of this paper must be obtained from the Editor. 1977 studies the speech database consisted of utterances recorded under laboratory conditions (i.e., cooperative subjects, soundproof booths, and subject prompting) using dialed-up lines over a local Private Branch Exchange (PBX). Peak signal-to-noise ratios ranged from 40 to 60 dB under these conditions. The recognition systems previously studied involved either a user training phase (speaker-dependent systems) or no training phase (speaker-independent systems). The vocabulary sizes ranged from as few as 10 words,2 to as many as 1109 words.7 Our past studies of speaker-independent systems involved a relatively small number of subjects, typically 100 for training and 10 to 40 for test calculations. Our current recognition systems performed very well given these conditions.