PLDA-Based Diarization of Telephone Conversations

摘要

This paper investigates the application of the probabilistic lineardiscriminant analysis (PLDA) to speaker diarization of telephone conversations.We introduce using a variational Bayes (VB) approach for inference under a PLDAmodel for modeling segmental i-vectors in speaker diarization. Deterministicannealing (DA) algorithm is imposed in order to avoid local optimal solutionsin VB iterations. We compare our proposed system with a well-known system thatapplies k-means clustering on principal component analysis (PCA) coefficientsof segmental i-vectors. We used summed channel telephone data from the NationalInstitute of Standards and Technology (NIST) 2008 Speaker RecognitionEvaluation (SRE) as the test set in order to evaluate the performance of theproposed system. We achieve about 20% relative improvement in Diarization ErrorRate (DER) compared to the baseline system.