Damage Control During Domain Adaptation for Transducer Based Automatic Speech Recognition

TL;DR

This paper proposes techniques to adapt speech recognition models to new domains while preventing performance loss on original domains, without needing access to original training data.

Contribution

It introduces regularized adapter modules and limited training strategies for Transducer models to achieve domain adaptation with minimal forgetting.

Findings

Effective adaptation to new speech domains demonstrated

Significant reduction in degradation on original domain

Strong results on Google Speech Commands and UK dialect datasets

Abstract

Automatic speech recognition models are often adapted to improve their accuracy in a new domain. A potential drawback of model adaptation to new domains is catastrophic forgetting, where the Word Error Rate on the original domain is significantly degraded. This paper addresses the situation when we want to simultaneously adapt automatic speech recognition models to a new domain and limit the degradation of accuracy on the original domain without access to the original training dataset. We propose several techniques such as a limited training strategy and regularized adapter modules for the Transducer encoder, prediction, and joiner network. We apply these methods to the Google Speech Commands and to the UK and Ireland English Dialect speech data set and obtain strong results on the new target domain while limiting the degradation on the original domain.