RobustDistiller: Compressing Universal Speech Representations for Enhanced Environment Robustness

TL;DR

RobustDistiller is a novel knowledge distillation approach that compresses large universal speech representations into smaller, more robust models, significantly improving environmental noise and reverberation robustness for edge applications.

Contribution

It introduces a multi-task layer-wise distillation method that enhances robustness of compressed speech models against environmental artifacts.

Findings

Outperforms benchmarks across various noise types and reverberation levels.

Student model achieves comparable performance to larger teacher model.

Effective on multiple universal representations and downstream tasks.

Abstract

Self-supervised speech pre-training enables deep neural network models to capture meaningful and disentangled factors from raw waveform signals. The learned universal speech representations can then be used across numerous downstream tasks. These representations, however, are sensitive to distribution shifts caused by environmental factors, such as noise and/or room reverberation. Their large sizes, in turn, make them unfeasible for edge applications. In this work, we propose a knowledge distillation methodology termed RobustDistiller which compresses universal representations while making them more robust against environmental artifacts via a multi-task learning objective. The proposed layer-wise distillation recipe is evaluated on top of three well-established universal representations, as well as with three downstream tasks. Experimental results show the proposed methodology applied on top of the WavLM Base+ teacher model outperforming all other benchmarks across noise types and levels, as well as reverberation times. Oftentimes, the obtained results with the student model (24M parameters) achieved results inline with those of the teacher model (95M).