Deep Neural Networks for extracting Baum-Welch statistics for Speaker Recognition
Patrick J Kenny, Themos Stafylakis, Pierre Ouellet, Vishwa Gupta, Jahangir Alam · 2014
We examine the use of Deep Neural Networks (DNN) in extracting Baum-Welch statistics for i-vector-based text-independent speaker recognition. Instead of training the uni-versal background model using the standard EM algorithm, the components are predefined and correspond to the set of triphone states, the posterior occupancy probabilities of which are mod-eled by a DNN. Those assignments are then combined with the standard 60-dim MFCC features to calculate first order Baum-Welch statistics in order to train the i-vector extractor and ex-tract i-vectors. The DNN-based assignment force the i-vectors to capture the idiosyncratic way in which each speaker pro-nounces each particular triphone state, which can enrich the standard short-term spectral representation of the standard i-vectors. After experimenting with Switchboard data and a baseline PLDA classifier, our results showed that although the proposed i-vectors yield inferior performance compared to the standard ones, they are capable of attaining 16 % relative improvement when fused with them, meaning that they carry useful com-plementary information about the speaker’s identity. A further experiment with a different DNN configuration attained com-parable performance with the baseline i-vectors on NIST 2012 (condition C2, female). 1.