iVector Fusion of Prosodic and Cepstral Features for Speaker Verification

speaker verification, prosody, JFA, iVector, SMM, fusion

This publication is about the first results on the use of total variability modeling of the mean supervector space for a set of prosodic features. We show that this iVector approach outperforms the standard JFA approach originally proposed for these features. We note that this improvement over JFA is observed only when the iVectors are modeled using the PLDA back end.

In this paper we apply the promising iVector extraction technique followed by PLDA modeling to simple prosodic contour features. With this procedure we achieve results comparable to a system that models much more complex prosodic features using our recently proposed SMM-based iVector modeling technique. We then propose a combination of both prosodic iVectors by joint PLDA modeling that leads to significant improvements over individual systems with an EER of 5.4% on NIST SRE 2008 telephone data. Finally, we can combine these two prosodic iVector front ends with a baseline cepstral iVector system to achieve up to 21% relative reduction in new DCF.