Analysis of ABC Frontend Audio Systems for the NIST-SRE24

TL;DR

This paper analyzes various embedding extractors for speaker recognition in NIST SRE 2024, comparing architectures and training conditions to develop robust, state-of-the-art frontends for conversational telephone speech.

Contribution

It introduces a comprehensive evaluation of ResNet, ReDimNet, and XLS-R architectures trained under fixed and open conditions for speaker embedding extraction.

Findings

VoxBlink-trained models show high robustness across languages.

ResNet and ReDimNet architectures perform competitively.

Open condition training with VoxBlink improves performance.

Abstract

We present a comprehensive analysis of the embedding extractors (frontends) developed by the ABC team for the audio track of NIST SRE 2024. We follow the two scenarios imposed by NIST: using only a provided set of telephone recordings for training (fixed) or adding publicly available data (open condition). Under these constraints, we develop the best possible speaker embedding extractors for the pre-dominant conversational telephone speech (CTS) domain. We explored architectures based on ResNet with different pooling mechanisms, recently introduced ReDimNet architecture, as well as a system based on the XLS-R model, which represents the family of large pre-trained self-supervised models. In open condition, we train on VoxBlink2 dataset, containing 110 thousand speakers across multiple languages. We observed a good performance and robustness of VoxBlink-trained models, and our experiments show practical recipes for developing state-of-the-art frontends for speaker recognition.