Aligned Contrastive Predictive Coding

TL;DR

This paper introduces Aligned Contrastive Predictive Coding (ACPC), a self-supervised learning approach that encourages models to learn slowly varying, piece-wise constant latent representations, improving speech coding performance and training efficiency.

Contribution

The paper proposes ACPC, a novel contrastive predictive coding method that aligns predictions to produce stable, slowly varying latent representations for speech processing.

Findings

Higher linear phone prediction accuracy

Lower ABX error rates

Faster training due to fewer prediction heads

Abstract

We investigate the possibility of forcing a self-supervised model trained using a contrastive predictive loss to extract slowly varying latent representations. Rather than producing individual predictions for each of the future representations, the model emits a sequence of predictions shorter than that of the upcoming representations to which they will be aligned. In this way, the prediction network solves a simpler task of predicting the next symbols, but not their exact timing, while the encoding network is trained to produce piece-wise constant latent codes. We evaluate the model on a speech coding task and demonstrate that the proposed Aligned Contrastive Predictive Coding (ACPC) leads to higher linear phone prediction accuracy and lower ABX error rates, while being slightly faster to train due to the reduced number of prediction heads.