Getting More for Less: Using Weak Labels and AV-Mixup for Robust Audio-Visual Speaker Verification

TL;DR

This work explored multitask learning techniques to further enhance DML, and shows that an auxiliary task with even weak labels can increase the quality of the learned speaker representation without increasing model complexity during inference.

Abstract

Distance Metric Learning (DML) has typically dominated the audio-visual speaker verification problem space, owing to strong performance in new and unseen classes. In our work, we explored multitask learning techniques to further enhance DML, and show that an auxiliary task with even weak labels can increase the quality of the learned speaker representation without increasing model complexity during inference. We also extend the Generalized End-to-End Loss (GE2E) to multimodal inputs and demonstrate that it can achieve competitive performance in an audio-visual space. Finally, we introduce AV-Mixup, a multimodal augmentation technique during training time that has shown to reduce speaker overfit. Our network achieves state of the art performance for speaker verification, reporting 0.244%, 0.252%, 0.441% Equal Error Rate (EER) on the VoxCeleb1-O/E/H test sets, which is to our knowledge, the best published results on VoxCeleb1-E and VoxCeleb1-H.

Figures (2)

• Figure 1: System level diagram of the REPTAR network. The multimodal representation feeds into the training heads, which are removed during inference time. For inference, the multimodal representation can be used directly for speaker verification
• Figure 2: Interclass and Intraclass Pairwise Distance Distribution of Triplet and GE2E-MM models for a random speaker.