Self-supervised contrastive speaker verification with nearest neighbor positive instances

Yan Liu, Li-Fang Wei, Chuan-Fei Zhang, Tian-Hao Zhang, Song-Lu Chen, Xu-Cheng Yin

Abstract: Highlights•We use nearest neighbor positive instances selected from a dynamic queue to improve the positive instances diversity.•Our self-supervised contrastive training model achieves competitive performance compared to previous models.•Our self-supervised contrastive training model is better than supervised training models in cross-dataset testing.