Distilling HuBERT with LSTMs via Decoupled Knowledge Distillation

TL;DR

This paper demonstrates that decoupled knowledge distillation can effectively compress HuBERT into an LSTM-based model, reducing parameters while improving speech recognition performance.

Contribution

It introduces a novel approach to distilling HuBERT into an LSTM model using decoupled knowledge distillation, enabling flexible architecture and better efficiency.

Findings

LSTM-based model outperforms DistilHuBERT in speech recognition.

Parameter count is reduced below previous distilled models.

Improved recognition accuracy with fewer parameters.

Abstract

Much research effort is being applied to the task of compressing the knowledge of self-supervised models, which are powerful, yet large and memory consuming. In this work, we show that the original method of knowledge distillation (and its more recently proposed extension, decoupled knowledge distillation) can be applied to the task of distilling HuBERT. In contrast to methods that focus on distilling internal features, this allows for more freedom in the network architecture of the compressed model. We thus propose to distill HuBERT's Transformer layers into an LSTM-based distilled model that reduces the number of parameters even below DistilHuBERT and at the same time shows improved performance in automatic speech recognition.