Task Arithmetic can Mitigate Synthetic-to-Real Gap in Automatic Speech Recognition

TL;DR

This paper introduces SYN2REAL task vector arithmetic to reduce the performance gap between synthetic and real speech data in ASR models, significantly improving accuracy across multiple domains.

Contribution

The paper proposes a novel task vector arithmetic method, SYN2REAL, to effectively mitigate the synthetic-to-real gap in speech recognition models.

Findings

10.03% average WER improvement over baselines

Effective adaptation to multiple real speech domains

Task vectors enhance model robustness across domains

Abstract

Synthetic data is widely used in speech recognition due to the availability of text-to-speech models, which facilitate adapting models to previously unseen text domains. However, existing methods suffer in performance when they fine-tune an automatic speech recognition (ASR) model on synthetic data as they suffer from the distributional shift commonly referred to as the synthetic-to-real gap. In this paper, we find that task vector arithmetic is effective at mitigating this gap. Our proposed method, SYN2REAL task vector, shows an average improvement of 10.03\% improvement in word error rate over baselines on the SLURP dataset. Additionally, we show that an average of SYN2REAL task vectors, when we have real speeches from multiple different domains, can further adapt the original ASR model to perform better on the target text domain.