Towards Rehearsal-Free Multilingual ASR: A LoRA-based Case Study on Whisper

TL;DR

This paper explores LoRA-based strategies to adapt multilingual speech models like Whisper to new languages without retraining from scratch, aiming to prevent forgetting and improve efficiency.

Contribution

It introduces a novel approach using original LoRA parameters and a learnable rank coefficient to enhance multilingual model adaptation while preserving previous performance.

Findings

Better adaptation results with fewer parameters

Effective mitigation of catastrophic forgetting

Improved efficiency in multilingual speech recognition

Abstract

Pre-trained multilingual speech foundation models, like Whisper, have shown impressive performance across different languages. However, adapting these models to new or specific languages is computationally extensive and faces catastrophic forgetting problems. Addressing these issues, our study investigates strategies to enhance the model on new languages in the absence of original training data, while also preserving the established performance on the original languages. Specifically, we first compare various LoRA-based methods to find out their vulnerability to forgetting. To mitigate this issue, we propose to leverage the LoRA parameters from the original model for approximate orthogonal gradient descent on the new samples. Additionally, we also introduce a learnable rank coefficient to allocate trainable parameters for more efficient training. Our experiments with a Chinese Whisper model (for Uyghur and Tibetan) yield better results with a more compact parameter set.