Can Large Language Models Reliably Correct Errors in Low-Resource ASR? A Contamination-Aware Case Study on West Frisian
This work addresses the underexplored effectiveness of LLM-based error correction for low-resource ASR, while controlling for data contamination, which is a critical methodological concern for the speech community.
The study investigates whether large language models (LLMs) can reliably correct errors in low-resource ASR for West Frisian, finding that generative error correction (GER) improves ASR performance in most settings, with the best GPT-5.1 results surpassing oracle word error rates (WERs).
Automatic speech recognition (ASR) has improved substantially in recent years, yet performance remains limited for low-resource languages. Large language models (LLMs) have shown promise for improving ASR through generative error correction (GER), but their effectiveness in low-resource settings remains underexplored. In addition, it remains unclear to what extent data contamination influences the reported improvements in LLM-based GER. This study investigates LLM-based GER for low-resource Frisian. In addition to a public corpus, we construct and use a Frisian offline dataset with non-public texts for evaluation to control for potential data contamination. Results show that GER improves ASR performance in most settings, with the best GPT-5.1 results surpassing oracle WERs. Comparable gains on the offline dataset indicate that improvements reflect true correction ability. We further provide a detailed error analysis revealing model correction patterns.