Improving Contextual ASR via Multi-grained Fusion with Large Language Models

TL;DR

This paper introduces a multi-grained fusion method combining token-level and phrase-level techniques with Large Language Models to improve contextual keyword recognition in end-to-end ASR systems, achieving state-of-the-art results.

Contribution

It proposes a novel late-fusion approach that jointly leverages token and phrase-level information with LLMs for enhanced ASR performance.

Findings

Achieves state-of-the-art keyword recognition metrics on Chinese and English datasets.

Both token-level and phrase-level components significantly improve performance.

The joint multi-grained framework balances fine-grained accuracy with holistic understanding.

Abstract

While end-to-end Automatic Speech Recognition (ASR) models have shown impressive performance in transcribing general speech, they often struggle to accurately recognize contextually relevant keywords, such as proper nouns or user-specific entities. Previous approaches have explored leveraging keyword dictionaries in the textual modality to improve keyword recognition, either through token-level fusion that guides token-by-token generation or phrase-level fusion that enables direct copying of keyword phrases. However, these methods operate at different granularities and have their own limitations. In this paper, we propose a novel multi-grained fusion approach that jointly leverages the strengths of both token-level and phrase-level fusion with Large Language Models (LLMs). Our approach incorporates a late-fusion strategy that elegantly combines ASR's acoustic information with LLM's rich contextual knowledge, balancing fine-grained token precision with holistic phrase-level understanding. Experiments on Chinese and English datasets demonstrate that our approach achieves state-of-the-art performance on keyword-related metrics while preserving high accuracy on non-keyword text. Ablation studies further confirm that the token-level and phrase-level components both contribute significantly to the performance gains, complementing each other in our joint multi-grained framework. The code and models will be publicly available at https://github.com/.