Self-Supervised Frameworks for Speaker Verification via Bootstrapped Positive Sampling

TL;DR

This paper introduces Self-Supervised Positive Sampling (SSPS), a bootstrapped positive sampling method for SSL in speaker verification, which improves performance and robustness by selecting diverse positives close to anchors in representation space.

Contribution

The paper proposes SSPS, a novel bootstrapped positive sampling strategy that enhances SSL frameworks for speaker verification by reducing channel bias and intra-class variance.

Findings

SSPS improves SSL speaker verification performance on VoxCeleb benchmarks.

SimCLR with SSPS achieves 2.57% EER, comparable to DINO.

SSPS reduces intra-class variance and channel information in speaker representations.

Abstract

Recent developments in Self-Supervised Learning (SSL) have demonstrated significant potential for Speaker Verification (SV), but closing the performance gap with supervised systems remains an ongoing challenge. SSL frameworks rely on anchor-positive pairs, constructed from segments of the same audio utterance. Hence, positives have channel characteristics similar to those of their corresponding anchors, even with extensive data-augmentation. Therefore, this positive sampling strategy is a fundamental limitation as it encodes too much information regarding the recording source in the learned representations. This article introduces Self-Supervised Positive Sampling (SSPS), a bootstrapped technique for sampling appropriate and diverse positives in SSL frameworks for SV. SSPS samples positives close to their anchor in the representation space, assuming that these pseudo-positives belong to the same speaker identity but correspond to different recording conditions. This method consistently demonstrates improvements in SV performance on VoxCeleb benchmarks when applied to major SSL frameworks, including SimCLR, SwAV, VICReg, and DINO. Using SSPS, SimCLR and DINO achieve 2.57% and 2.53% EER on VoxCeleb1-O, respectively. SimCLR yields a 58% relative reduction in EER, getting comparable performance to DINO with a simpler training framework. Furthermore, SSPS lowers intra-class variance and reduces channel information in speaker representations while exhibiting greater robustness without data-augmentation.