Evaluation of an Automatic Speaker-Verification System Over Telephone Lines

Speaker verification is the authentication of an individual's claimed identity by analysis of his spoken utterances. Research on an automatic system for speaker verification at Bell Laboratories has been reported in previous papers. 1-3 The system is based on an acoustic analysis of a fixed, sentence-long utterance resulting in a function of time or contour for each feature analyzed. Features selected for analysis in previous evaluations have included pitch, intensity, the first three formants, and selected predictor coefficients. The system compares the set of sample contours obtained from an unknown individual with the set of reference contours corresponding to the identity claimed by that individual. If the comparison results in an overall measure of dissimilarity which is smaller than a predetermined threshold, the identity claim is accepted. Otherwise, it is rejected. 723 Previous evaluations of the system have concentrated on investigating features to be analyzed and developing comparison procedures to make the system as effective as possible in terms of reducing overall error rate. The speech samples used in these evaluations were collected by wideband recording of male speakers in a sound booth. The recordings were selected and edited to eliminate botched utterances and nonspeech acoustic events. From the outset, however, the intent has been to provide a completely automatic system which could operate via dialed-up lines from telephones on the user's own premises and to include both male and female speakers in the user population.