Augmentation Adversarial Training for Self-Supervised Speaker Representation Learning

Jingu Kang

doi:10.1109/JSTSP.2022.3200915

Augmentation Adversarial Training for Self-Supervised Speaker Representation Learning

2022-10-01

9

PRE

AI

DOI：10.1109/JSTSP.2022.3200915

Original

Original request for help

Save

Abstract

En 中文

The goal of this work is to train robust speaker recognition models using self-supervised representation learning. Recent works on self-supervised speaker representations are based on contrastive learning in which they encourage within-utterance embeddings to be similar and across-utterance embeddings to be dissimilar. However, since the within-utterance segments share the same acoustic characteristics, it is difficult to separate the speaker information from the channel information. To this end, we propose an augmentation adversarial training strategy that trains the network to be discriminative for the speaker information, while invariant to the augmentation applied. Since the augmentation simulates the acoustic characteristics, training the network to be invariant to augmentation also encourages the network to be invariant to the channel information in general. Extensive experiments on the VoxCeleb and VOiCES datasets show significant improvements over previous works using self-supervision, and the performance of our self-supervised models far exceeds that of humans. We also conduct semi-supervised learning experiments to show that augmentation adversarial training benefits performance in presence of speaker labels.

Keywords:

Training

Speaker recognition

Measurement

Task analysis

Semisupervised learning

Representation learning

Entropy

Self-supervised learning

speaker recognition

Augmentation Adversarial Training for Self-Supervised Speaker Representation Learning

Abstract

Journal

Organization