DNN-Driven Mixture of PLDA for Robust Speaker Verification

Abstract

The mismatch between enrollment and test utterances due to different types of variabilities is a great challenge in speaker verification. Based on the observation that the SNRlevel variability or channel-type variability causes heterogeneous clusters in i-vector space, this paper proposes to apply supervised learning to drive or guide the learning of PLDA mixture models. Specifically, a deep neural network (DNN) is trained to produce the posterior probabilities of different SNR levels or channel types given i-vectors as input. These posteriors then replace the posterior probabilities of indicator variables in the mixture of PLDA. The discriminative training causes the mixture model to perform more reasonable soft divisions of the i-vector space as compared to the conventional mixture of PLDA. During verification, given a test i-vector and a targetspeaker's i-vector, the marginal likelihood for the same-speaker hypothesis is obtained by summing the component likelihoods weighted by the component posteriors produced by the DNN, and likewise for the different-speaker hypothesis. Results based on NIST 2012 SRE demonstrate that the proposed scheme leads to better performance under more realistic situations where both training and test utterances cover a wide range of SNRs and different channel types. Unlike the previous SNR-dependent mixture of PLDA which only focuses on SNR mismatch, the proposed model is more general and is potentially applicable to addressing different types of variability in speech.

i-Vector DNN Scoring and Calibration for Noise Robust Speaker Verification

“…In [15], Mak et al advocated that utterances of different SNR levels will not only cause the i-vectors to fall on different regions of the i-vector spaces but also change the orientation of the speaker subspace. A mixture PLDA model with mixture alignments determined by the SNR level of utterances [15] or by their i-vectors [16] was then derived to model the SNR-dependent i-vectors.…”

Edge-centric multimodal authentication system using encrypted biometric templates

“…The identity of a person can be verified in the biometric authentication systems using personal attributes, such as speech [16,17], face [18,19], fingerprints [20,21], palmprint [22,23], gait [24,25], and iris [26,27]. These physiological and behavioral attributes of humans are more reliable in authentication compared with knowledge-based or token-based approaches because these attributes cannot be stolen and are unique for every individual.…”

Denoised Senone I-Vectors for Robust Speaker Verification

“…For example, Hasan et al [11] and Garcia-Romero et al [12] trained a PLDA model by pooling speeches from multiple conditions, and Li and Mak [13], [14] modeled the noise-level variability in utterances by introducing an SNR factor and an SNR subspace into the PLDA model. In [15], [16], Mak et al advocated that utterances of different SNR levels will not only cause i-vectors to fall on different regions of the i-vector spaces but also change the orientation of the speaker subspace. A mixture PLDA model with mixture alignments determined by the SNR level of utterances was then derived to model SNR-dependent i-vectors.…”