Cluster-dependent feature transformation with divergence-based out-of-handset rejection for robust speaker verification

This paper proposes a divergence-based cluster selector with out-of-handset (OOH) rejection capability to identify the ´unseen´ handsets. This is achieved by measuring the Jensen difference between the selector´s output and a constant vector with identical elements. The resulting cluster selector is combined with a feature-based channel compensation algorithm for telephone-based speaker verification. Utterances whose handsets are identified as ´unseen´ are normalized by cepstral mean subtraction (CMS). On the other hand, if the handset can be identified (considered as ´seen´), a corresponding set of cluster-dependent transformation parameters are used to transform the utterances. Experiments based on ten handsets of the HTIMIT corpus show that using the cluster-dependent transformation parameters to transform the utterances with correctly identified handsets and processing those utterances with ´unseen´ handsets by CMS achieve the best result.