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

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xu, Tianyi, Huang, Kaixun, Guo, Pengcheng, Zhou, Yu, Huang, Longtao, Xue, Hui, Xie, Lei
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916362895491072
author Xu, Tianyi
Huang, Kaixun
Guo, Pengcheng
Zhou, Yu
Huang, Longtao
Xue, Hui
Xie, Lei
author_facet Xu, Tianyi
Huang, Kaixun
Guo, Pengcheng
Zhou, Yu
Huang, Longtao
Xue, Hui
Xie, Lei
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2408_10680
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Towards Rehearsal-Free Multilingual ASR: A LoRA-based Case Study on Whisper
Xu, Tianyi
Huang, Kaixun
Guo, Pengcheng
Zhou, Yu
Huang, Longtao
Xue, Hui
Xie, Lei
Computation and Language
Sound
Audio and Speech Processing
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.
title Towards Rehearsal-Free Multilingual ASR: A LoRA-based Case Study on Whisper
topic Computation and Language
Sound
Audio and Speech Processing
url https://arxiv.org/abs/2408.10680