Confidence Bounds Curves as a Tool for Evaluation of Automatic Speaker Recognition Results Uncertainty
In some applications of speaker recognition, for example in the forensic area or in the access control systems, an important task is to estimate some absolute measure of identity of the speakers. Automatic speaker recognition methods in this case seem to be the fastest and the simplest speaker identification tool [1-2]. However, up to now the applicability and reliability evaluation of automatic speaker recognition systems (ASRS) for single cases, e.g. in forensic area, is widely disputable [3-7]. Output results of state-of-the-art ASRS are based on statistical data analysis. Their applicability for individual comparisons is theoretically and practically rather complicated task. In this paper we address the issue of more detailed analysis of training data statistical structure and more careful decision making for ASRS in the context of one-to-one speech recordings comparisons using confidence bounds curve idea and bootstrap calculating technique.