Bayesian Learning for Domain-Invariant Speaker Verification and Anti-Spoofing

TL;DR

This paper introduces Bayesian weighted RFN (BWRFN), a novel method using variational inference to adapt frequency weights in speaker verification and anti-spoofing, improving robustness under domain mismatch conditions.

Contribution

It proposes a Bayesian approach to frequency normalization, modeling weight uncertainty to enhance domain invariance in speaker verification and anti-spoofing systems.

Findings

BWRFN outperforms RFN and WRFN in cross-dataset ASV tasks.

BWRFN improves anti-spoofing robustness against domain shifts.

Bayesian modeling effectively captures weight uncertainty, enhancing performance.

Abstract

The performance of automatic speaker verification (ASV) and anti-spoofing drops seriously under real-world domain mismatch conditions. The relaxed instance frequency-wise normalization (RFN), which normalizes the frequency components based on the feature statistics along the time and channel axes, is a promising approach to reducing the domain dependence in the feature maps of a speaker embedding network. We advocate that the different frequencies should receive different weights and that the weights' uncertainty due to domain shift should be accounted for. To these ends, we propose leveraging variational inference to model the posterior distribution of the weights, which results in Bayesian weighted RFN (BWRFN). This approach overcomes the limitations of fixed-weight RFN, making it more effective under domain mismatch conditions. Extensive experiments on cross-dataset ASV, cross-TTS anti-spoofing, and spoofing-robust ASV show that BWRFN is significantly better than WRFN and RFN.