Mixture of LoRA Experts for Low-Resourced Multi-Accent Automatic Speech Recognition

Raphaël, Irina, Emmanuel · 2025

Read Mixture of LoRA Experts for Low-Resourced Multi-Accent Automatic Speech Recognition by Raphaël, Irina, Emmanuel (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-20006.txt.

Description

Raphaël Bagat, Irina Illina, and Emmanuel Vincent improved speech recognition for non-native accents in two thousand twenty-five. Systems often fail on accented speech, and data for many accents are scarce. Their method fine-tunes a model with a mixture of small adapter experts, each tuned to one accent.

The method works whether the accent is known at use time or not, with no need to retrain. Tests used the Whisper model on a corpus of second-language English speech. Word error fell clearly versus standard adapters and full fine-tuning when the accent was unknown, and fell further when it was known. The mixture also forgot less of its prior skill than rival methods.

Catalog record

CREATORRaphaël, Irina, Emmanuel
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-20006

Browse the full library · About the Tower of Babel