GLoRIA: Gated Low-Rank Interpretable Adaptation for Dialectal ASR
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912938904780800 |
|---|---|
| author | Mehralian, Pouya Farasyn, Melissa Breitbarth, Anne Ghyselen, Anne-Sophie Van hamme, Hugo |
| author_facet | Mehralian, Pouya Farasyn, Melissa Breitbarth, Anne Ghyselen, Anne-Sophie Van hamme, Hugo |
| contents | Automatic Speech Recognition (ASR) in dialect-heavy settings remains challenging due to strong regional variation and limited labeled data. We propose GLoRIA, a parameter-efficient adaptation framework that leverages metadata (e.g., coordinates) to modulate low-rank updates in a pre-trained encoder. GLoRIA injects low-rank matrices into each feed-forward layer, with a gating MLP determining the non-negative contribution of each LoRA rank-1 component based on location metadata. On the GCND corpus, GLoRIA outperforms geo-conditioned full fine-tuning, LoRA, and both dialect-specific and unified full fine-tuning, achieving state-of-the-art word error rates while updating under 10% of parameters. GLoRIA also generalizes well to unseen dialects, including in extrapolation scenarios, and enables interpretable adaptation patterns that can be visualized geospatially. These results show metadata-gated low-rank adaptation is an effective, interpretable, and efficient solution for dialectal ASR. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_02464 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | GLoRIA: Gated Low-Rank Interpretable Adaptation for Dialectal ASR Mehralian, Pouya Farasyn, Melissa Breitbarth, Anne Ghyselen, Anne-Sophie Van hamme, Hugo Computation and Language Artificial Intelligence Automatic Speech Recognition (ASR) in dialect-heavy settings remains challenging due to strong regional variation and limited labeled data. We propose GLoRIA, a parameter-efficient adaptation framework that leverages metadata (e.g., coordinates) to modulate low-rank updates in a pre-trained encoder. GLoRIA injects low-rank matrices into each feed-forward layer, with a gating MLP determining the non-negative contribution of each LoRA rank-1 component based on location metadata. On the GCND corpus, GLoRIA outperforms geo-conditioned full fine-tuning, LoRA, and both dialect-specific and unified full fine-tuning, achieving state-of-the-art word error rates while updating under 10% of parameters. GLoRIA also generalizes well to unseen dialects, including in extrapolation scenarios, and enables interpretable adaptation patterns that can be visualized geospatially. These results show metadata-gated low-rank adaptation is an effective, interpretable, and efficient solution for dialectal ASR. |
| title | GLoRIA: Gated Low-Rank Interpretable Adaptation for Dialectal ASR |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2603.02464 |