Fact · Voice and AI
The 2025 INTERSPEECH authors add the speech recognizer's N-best hypotheses (its ranked alternative transcripts) and phonetic context to the language model, to reduce over-correction. They report lower word and character error rates on English and Japanese datasets.
Checked against its source on October 17, 2026
In the source's own words
“we integrate ASR's N-best hypotheses along with phonetic context to mitigate over-correction.”
LLM-based Generative Error Correction for Rare Words with Synthetic Data and Phonetic Context, Natsuo Yamashita et al., arXiv (accepted by INTERSPEECH 2025), 2025, Abstract.
Open the original- What it measures
- The method targets rare words. Its fine-tuning uses synthetic data that contains rare words, and it reports gains on both rare-word correction and overall error rates.
- What it does not show
- Reported on English and Japanese test sets chosen for rare words and domain vocabulary. The abstract does not say how often the correction step changes correct words in everyday dictation.