How to Recognize New Words: A Comparison Between Context Biasing Methods and Speech LLMs

2026-08-06

AI Summary

This study addresses the performance degradation of automatic speech recognition (ASR) systems on out-of-vocabulary terms—such as proper nouns, acronyms, and domain-specific terminology—when training data is scarce. The authors systematically compare contextual biasing techniques with speech large language models (Speech LLMs) in both read and spontaneous speech scenarios, evaluating word error rate (WER) on target words. Experiments leverage Whisper-based contextual biasing and three Speech LLMs. Results demonstrate that contextual biasing reduces target-word WER by up to 88% without adversely affecting other vocabulary. While Speech LLMs excel in read speech, they exhibit limited generalization in spontaneous settings and are sensitive to the number of distractors and prompt ordering. This work delineates the operational boundaries and performance trade-offs between these two approaches.

Abstract

Recognizing new and rare words - named entities, acronyms, domain specific special words, and other items scarce in training data - remains a key challenge for automatic speech recognition (ASR). We compare two strategies for this: context biasing methods, where an ASR model is extended such that during inference a word list can be supplied, and speech large language models (LLMs) prompted with context directly. We evaluate two context biasing methods based on Whisper against three speech LLMs across read and non-read speech, reporting biased, unbiased, and overall word error rate (WER). The context biasing methods cut biased WER by up to 88% relative while leaving other words largely unaffected. Speech LLMs excel on read speech but generalize less well to non-read speech, and prove sensitive to distractor count and prompt word order. We characterize the resulting trade-offs to guide method selection.

Problem

• automatic speech recognition
• new words
• rare words
• named entities
• domain-specific terms

Innovation

• context biasing
• speech LLMs
• new word recognition
• automatic speech recognition
• word error rate