In this paper we propose a framework for measuring the overalludperformance of an automatic speaker recognition system usinguda set of trials of a heterogeneous evaluation such as NIST SRE-ud2008, which combines several acoustic conditions in one evalu-udation. We do this by weighting trials of different conditions ac-udcording to their relative proportion, and we derive expressionsudfor the basic speaker recognition performance measures Cdet,udCllr, as well as the DET curve, from which EER and Cmin can detudbe computed. Examples of pooling of conditions are shown on SRE-2008 data, including speaker sex and microphone type and speaking style.
展开▼