Effective speaker spotting for watch‐list detection of fraudsters in telephone banking

Abstract

This study describes a special-case application of speaker recognition in open-set speaker-identification mode, which nonetheless has wide applicability. Watch-list based speaker spotting in telephone banking can potentially provide valuable protection against 'known' fraudsters with access to stolen customer details. In this study, the detection of known fraudsters in a telephone banking service using commercial off-the-shelf verification engines is described. A new 'delta scoring' method for watch-list detection is proposed based on using the genuine customer model as a reference. The approach combines for the first time speaker recognition in both verification and identification mode. Empirical experiment results show a significant gain in performance using the new method.

MCE 2018: The 1st Multi-Target Speaker Detection and Identification Challenge Evaluation

“…As expected, the Top-1 stack detector performs worse than the top-S stack detector as the number of blacklist speakers increases. 5 Available in https://github.com/swshon/multi-speakerID This severe performance degradation could be a major issue when handling large-scale multi-target detection. Thus, in this challenge, we included a large blacklist set to assess how well current speech technology is able to detect and identify blacklists, and to explore algorithms incorporating a speaker representation such as an i-vector.…”

“…Since we did not receive any open condition submission, only fixed condition results are described in this paper. The baseline system 5 is based on the multi-target detector in [1]. For each input, we rank the multi-target detector scores and accept the top-k hypotheses if the rank-1 score is above a detection threshold.…”

“…Multitarget detection such as blacklist or watchlist was often described as open-set speaker identification. There are a few relevant studies [2,3,4,5,6,7], but it is not actively being explored because it is regarded as a special case of speaker verification. Most research on this topic pre-date the i-vector [8], so it is difficult to compare the performance of older blacklist detection systems with state-of-the-art technology.…”