Learn and Don't Forget: Adding a New Language to ASR Foundation Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Qian, Mengjie, Tang, Siyuan, Ma, Rao, Knill, Kate M., Gales, Mark J. F.
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913515434934272
author Qian, Mengjie
Tang, Siyuan
Ma, Rao
Knill, Kate M.
Gales, Mark J. F.
author_facet Qian, Mengjie
Tang, Siyuan
Ma, Rao
Knill, Kate M.
Gales, Mark J. F.
contents Foundation ASR models often support many languages, e.g. 100 languages in Whisper. However, there has been limited work on integrating an additional, typically low-resource, language, while maintaining performance on the original language set. Fine-tuning, while simple, may degrade the accuracy of the original set. We compare three approaches that exploit adaptation parameters: soft language code tuning, train only the language code; soft prompt tuning, train prepended tokens; and LoRA where a small set of additional parameters are optimised. Elastic Weight Consolidation (EWC) offers an alternative compromise with the potential to maintain performance in specific target languages. Results show that direct fine-tuning yields the best performance for the new language but degrades existing language capabilities. EWC can address this issue for specific languages. If only adaptation parameters are used, the language capabilities are maintained but at the cost of performance in the new language.
format Preprint
id arxiv_https___arxiv_org_abs_2407_06800
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Learn and Don't Forget: Adding a New Language to ASR Foundation Models
Qian, Mengjie
Tang, Siyuan
Ma, Rao
Knill, Kate M.
Gales, Mark J. F.
Audio and Speech Processing
Computation and Language
Machine Learning
Sound
Foundation ASR models often support many languages, e.g. 100 languages in Whisper. However, there has been limited work on integrating an additional, typically low-resource, language, while maintaining performance on the original language set. Fine-tuning, while simple, may degrade the accuracy of the original set. We compare three approaches that exploit adaptation parameters: soft language code tuning, train only the language code; soft prompt tuning, train prepended tokens; and LoRA where a small set of additional parameters are optimised. Elastic Weight Consolidation (EWC) offers an alternative compromise with the potential to maintain performance in specific target languages. Results show that direct fine-tuning yields the best performance for the new language but degrades existing language capabilities. EWC can address this issue for specific languages. If only adaptation parameters are used, the language capabilities are maintained but at the cost of performance in the new language.
title Learn and Don't Forget: Adding a New Language to ASR Foundation Models
topic Audio and Speech Processing
Computation and Language
Machine Learning
Sound
url https://arxiv.org/abs/2407.06800