Reflection of Conditional Independence Structure to Noise Variability for Noise Robust Text Dependent Speaker Verification

In the field of speaker verification (SV), development noise-robust systems is a challenge for their deployment in real-world environments. Noise variability compensation common strategy increasing robustness to noise variations. The performance depends on how well variability, which inherent within-class estimated. However, date, there no information about true that could reduce gap between empirical and statistics. Most studies assume covariates are independent. This study aims demonstrate assumption has conditional independence structure rather than an structure. was motivated by our previous findings, revealed optimal text-dependent (TD-SV) clean indicates optima all variabilities except structure; however, it unknown whether this also variability. Our supported experimental results obtained under noisy TD-SV trials using built with graphical least absolute shrinking selection operator-based probabilistic linear discriminant analysis, achieved up 10% relative equal error rate improvements.