Speaker Diarization and Automatic Speech Recognition have been a topic of research for decades. Evaluating the developed systems has been required for almost as long. Following the NIST initiatives a number of metrics have become standard to handle these evaluations, namely the Diarization Error Rate and the Word Error Rate. The initial definitions of these metrics and, more importantly, their ...

The novel approach to speaker adaptation within speech recognition system basing on late clustering of prototype speakers is presented. For a new speaker the speaker prototype is created dynamically on the basis of selected remembered prototypes that are similar enough to the new one. The training utterances are prepared in an optimized way to decrease training duration without negative influen...

The aim of this paper is to investigate the ways of interpreting evidence within the ®eld of speaker recognition. Several methods ± speaker veri®cation, speaker identi®cation and type I and type II errors statement ± will be presented and evaluated in the light of judicial needs. It will be shown that these methods for interpreting evidence unfortunately force the scientist to adopt a role and ...

We discuss the history and purposes of the NIST evaluations of speaker recognition performance. We cover the sites that have participated, the performance measures used, and the formats used to report results. We consider the extent to which there has been measurable progress over the years. In particular, we examine apparent performance improvements seen in the 2001 evaluation. Information for...

In the present work we discuss the results, which our speaker verification system, WCL-1, obtained in the 2003 NFI/TNO Forensic Speaker Recognition Evaluation. These results, together with the ones obtained in the 2003 NIST Speaker Recognition Evaluation, give opportunity for in depth analysis of the various aspects of real-world application of the speaker recognition technology. Based on the d...

The very large set of trials in the SRE10 extended evaluation [1] provides opportunity to study the effect of various factors on speaker recognition performance. This paper addresses the issue of age difference between target and non-target speakers and shows that false alarm probability is reduced substantially as the age difference increases. False alarm probability is significantly reduced f...

A professional impersonator has been studied when training his voice to mimic two target speakers. A three-fold investigation has been conducted; a computer-based speaker verification system was used, phonetic-acoustic measurements were made and a perception test was conducted. Our idea behind using this type of system is to measure how close to the target voice a professional impersonation mig...

Although many of the acoustic cues used for speaker identification change systematically with the voice level of the talker, little is known about the influence that vocal effort has on the identification of individual talkers by human listeners. In this experiment, listeners were trained to identify four different same-sex talkers speaking at one of three different levels of vocal effort (whis...

The so called Phone Log-Likelihood Ratio (PLLR) features, computed on phone posterior probabilities provided by phonetic decoders, convey acoustic-phonetic information in a sequence of frame-level vectors. Thus, PLLRs can be easily plugged into traditional acoustic systems just by replacing MFCCs, PLPs or whatever other representation. PLLR features were used under an iVector-PLDA approach in o...