Towards Supervised Performance on Speaker Verification with Self-Supervised Learning by Leveraging Large-Scale ASR Models
This work addresses the problem of improving speaker verification accuracy for applications like security and authentication by advancing self-supervised learning methods, though it is incremental as it builds on existing SSL and ASR model techniques.
This paper tackles the challenge of narrowing the performance gap between self-supervised and supervised speaker verification by proposing a framework that fine-tunes a pre-trained WavLM model with pseudo-labels derived from an SSL model, achieving 0.99% EER on VoxCeleb1-O, which is close to the supervised baseline of 0.94% EER.
Recent advancements in Self-Supervised Learning (SSL) have shown promising results in Speaker Verification (SV). However, narrowing the performance gap with supervised systems remains an ongoing challenge. Several studies have observed that speech representations from large-scale ASR models contain valuable speaker information. This work explores the limitations of fine-tuning these models for SV using an SSL contrastive objective in an end-to-end approach. Then, we propose a framework to learn speaker representations in an SSL context by fine-tuning a pre-trained WavLM with a supervised loss using pseudo-labels. Initial pseudo-labels are derived from an SSL DINO-based model and are iteratively refined by clustering the model embeddings. Our method achieves 0.99% EER on VoxCeleb1-O, establishing the new state-of-the-art on self-supervised SV. As this performance is close to our supervised baseline of 0.94% EER, this contribution is a step towards supervised performance on SV with SSL.