Dynamic ASR Pathways: An Adaptive Masking Approach Towards Efficient Pruning of A Multilingual ASR Model

TL;DR

This paper introduces Dynamic ASR Pathways, an adaptive masking method for efficient pruning of multilingual speech recognition models, enabling better performance and reduced need for language-specific pruning.

Contribution

It proposes a novel adaptive masking approach that dynamically adjusts sub-networks, improving pruning efficiency and performance in multilingual ASR models.

Findings

Outperforms existing pruning methods for sparse monolingual models.

Jointly discovers and trains better sub-networks in multilingual models.

Reduces the need for language-specific pruning processes.

Abstract

Neural network pruning offers an effective method for compressing a multilingual automatic speech recognition (ASR) model with minimal performance loss. However, it entails several rounds of pruning and re-training needed to be run for each language. In this work, we propose the use of an adaptive masking approach in two scenarios for pruning a multilingual ASR model efficiently, each resulting in sparse monolingual models or a sparse multilingual model (named as Dynamic ASR Pathways). Our approach dynamically adapts the sub-network, avoiding premature decisions about a fixed sub-network structure. We show that our approach outperforms existing pruning methods when targeting sparse monolingual models. Further, we illustrate that Dynamic ASR Pathways jointly discovers and trains better sub-networks (pathways) of a single multilingual model by adapting from different sub-network initializations, thereby reducing the need for language-specific pruning.