GLoRIA: Gated Low-Rank Interpretable Adaptation for Dialectal ASR

2026-03-02

AI Summary

This work addresses the performance bottleneck in dialect speech recognition caused by substantial regional variation and scarce labeled data. The authors propose a parameter-efficient adaptation framework that leverages metadata such as geographic location to construct a gating mechanism, which dynamically modulates the contributions of rank-1 components within LoRA modules inserted into a pretrained speech encoder. By updating fewer than 10% of the model parameters, the method achieves effective adaptation across multiple dialects. It offers both strong generalization and interpretability, attaining state-of-the-art word error rates on the GCND corpus and demonstrating robust performance on unseen dialects and in out-of-distribution extrapolation scenarios.

Technology Category

Machine Learning: Transfer, Domain Adaptation, Multi-Task LearningNatural Language Processing: SpeechSearch and Optimization: Learning to Search

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingUser Modeling, Personalization and Recommendation: Fairness-aware retrieval and rankingSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

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.

Problem

• dialectal ASR
• regional variation
• limited labeled data
• automatic speech recognition

Innovation

• low-rank adaptation
• metadata gating
• dialectal ASR
• parameter-efficient tuning
• interpretable adaptation