Vector-Quantized Autoregressive Predictive Coding
This work addresses a fundamental issue in self-supervised learning for speech processing, offering insights into representation learning, but it is incremental as it builds on existing APC methods.
The paper tackles the unclear link between self-supervised loss and downstream task performance in Autoregressive Predictive Coding (APC) by proposing Vector-Quantized APC (VQ-APC), which uses quantized representations to control information encoding; they find that at a specific model capacity, phonetic and speaker information are amplified, with learned codes corresponding well to English phones.
Autoregressive Predictive Coding (APC), as a self-supervised objective, has enjoyed success in learning representations from large amounts of unlabeled data, and the learned representations are rich for many downstream tasks. However, the connection between low self-supervised loss and strong performance in downstream tasks remains unclear. In this work, we propose Vector-Quantized Autoregressive Predictive Coding (VQ-APC), a novel model that produces quantized representations, allowing us to explicitly control the amount of information encoded in the representations. By studying a sequence of increasingly limited models, we reveal the constituents of the learned representations. In particular, we confirm the presence of information with probing tasks, while showing the absence of information with mutual information, uncovering the model's preference in preserving speech information as its capacity becomes constrained. We find that there exists a point where phonetic and speaker information are amplified to maximize a self-supervised objective. As a byproduct, the learned codes for a particular model capacity correspond well to English phones.