Telephony fraud detection using voice recognition techniques

Abstract

A system/method is disclosed for screening telephone system callers, using speaker-dependent text-independent voice recognition, to either prevent excessive fraudulent use of certain telephone system lines (or radio channels) or to selectively restrict classes of service allowed for different users of a common line. The voice recognition process is configured to provide an extent of mismatch indication, between a voice sample taken from a caller immediately using the protected line (or channel) and a recorded reference voice sample, which indication corresponds to the probability that the instant caller is not the same as the person who provided the reference sample. The extent of mismatch indication is compared to a mismatch limit value which can be set to a different magnitude for each call. Different actions are taken depending upon whether the extent of mismatch indication is or is not larger than the limit value. In situations screening for potential fraudulent use of a line/channel, a mismatch indication larger than the set limit is considered representative of possible fraud and further action is taken to either immediately block continuation of the call or to acquire additional information from the caller which can be used to verify authorized usage.