Towards Rehearsal-Free Multilingual ASR: A LoRA-based Case Study on Whisper
This addresses the challenge of efficient multilingual ASR adaptation for specific languages, though it is incremental as it builds on existing LoRA methods.
The study tackled the problem of adapting multilingual speech models like Whisper to new languages without access to original training data, while avoiding catastrophic forgetting, and achieved better results with a more compact parameter set in experiments on Chinese Whisper for Uyghur and Tibetan.
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.