SSPS: Self-Supervised Positive Sampling for Robust Self-Supervised Speaker Verification

TL;DR

This paper introduces SSPS, a novel positive sampling method for self-supervised speaker verification that improves performance by better capturing speaker identity across different recording conditions.

Contribution

The paper proposes Self-Supervised Positive Sampling (SSPS), a new technique that enhances SSL speaker verification by selecting positives based on clustering in latent space, reducing channel information bias.

Findings

SSPS improves speaker verification accuracy on VoxCeleb1-O.

SimCLR-SSPS achieves a 58% reduction in EER.

SSPS outperforms state-of-the-art SSL methods.

Abstract

Self-Supervised Learning (SSL) has led to considerable progress in Speaker Verification (SV). The standard framework uses same-utterance positive sampling and data-augmentation to generate anchor-positive pairs of the same speaker. This is a major limitation, as this strategy primarily encodes channel information from the recording condition, shared by the anchor and positive. We propose a new positive sampling technique to address this bottleneck: Self-Supervised Positive Sampling (SSPS). For a given anchor, SSPS aims to find an appropriate positive, i.e., of the same speaker identity but a different recording condition, in the latent space using clustering assignments and a memory queue of positive embeddings. SSPS improves SV performance for both SimCLR and DINO, reaching 2.57% and 2.53% EER, outperforming SOTA SSL methods on VoxCeleb1-O. In particular, SimCLR-SSPS achieves a 58% EER reduction by lowering intra-speaker variance, providing comparable performance to DINO-SSPS.