Contrastive Separative Coding for Self-supervised Representation Learning

TL;DR

This paper introduces Contrastive Separative Coding (CSC), a self-supervised learning method that extracts robust speech representations by separating target signals from interfering noise using a multi-task encoder, cross-attention, and a novel contrastive loss, improving speaker verification in noisy conditions.

Contribution

The paper proposes a novel self-supervised learning framework that focuses on separating signals from interference, with a new contrastive loss that does not require negative sampling, enhancing robustness in speech representation learning.

Findings

Achieves strong speaker verification performance in adverse conditions.

Introduces a negative-sampling-free contrastive loss.

Demonstrates effectiveness of cross-attention in separating signals.

Abstract

To extract robust deep representations from long sequential modeling of speech data, we propose a self-supervised learning approach, namely Contrastive Separative Coding (CSC). Our key finding is to learn such representations by separating the target signal from contrastive interfering signals. First, a multi-task separative encoder is built to extract shared separable and discriminative embedding; secondly, we propose a powerful cross-attention mechanism performed over speaker representations across various interfering conditions, allowing the model to focus on and globally aggregate the most critical information to answer the "query" (current bottom-up embedding) while paying less attention to interfering, noisy, or irrelevant parts; lastly, we form a new probabilistic contrastive loss which estimates and maximizes the mutual information between the representations and the global speaker vector. While most prior unsupervised methods have focused on predicting the future, neighboring, or missing samples, we take a different perspective of predicting the interfered samples. Moreover, our contrastive separative loss is free from negative sampling. The experiment demonstrates that our approach can learn useful representations achieving a strong speaker verification performance in adverse conditions.