LIA system description for NIST SRE 2016

TL;DR

This paper details the LIA speaker recognition system for NIST SRE 2016, combining eight sub-systems based on i-vector/PLDA with various features and data-shifting techniques, fused at score level.

Contribution

It introduces a multi-sub-system speaker recognition system utilizing diverse feature extraction and data-shifting methods, optimized for NIST SRE 2016 evaluation.

Findings

Achieved competitive speaker recognition performance.

Demonstrated effectiveness of fusion of multiple sub-systems.

Validated robustness of diverse feature and data-shifting combinations.

Abstract

This paper describes the LIA speaker recognition system developed for the Speaker Recognition Evaluation (SRE) campaign. Eight sub-systems are developed, all based on a state-of-the-art approach: i-vector/PLDA which represents the mainstream technique in text-independent speaker recognition. These sub-systems differ: on the acoustic feature extraction front-end (MFCC, PLP), at the i-vector extraction stage (UBM, DNN or two-feats posteriors) and finally on the data-shifting (IDVC, mean-shifting). The submitted system is a fusion at the score-level of these eight sub-systems.