Fact · Voice and AI
77.1percent relative reduction in word error rate
In the HyPoradise benchmark, the H2T-LoRA language model correction reduced word error rate by 77.1 percent relative on the ATIS test set and by 55.1 percent relative on WSJ.
Checked against its source on October 17, 2026
In the source's own words
“H2T-LoRA respectively reduces 77.1% and 55.1% relative WER on ATIS and WSJ”
HyPoradise: An Open Baseline for Generative Speech Recognition with Large Language Models (full text, HTML), Chen Chen et al., arXiv (NeurIPS 2023 Datasets and Benchmarks track), 2023, Results, Table 2.
Open the original- What it measures
- A large language model reads the recognizer's N-best hypotheses (its ranked alternative transcripts) and outputs a corrected transcript. H2T-LoRA is the variant adapted to each domain with LoRA, a lightweight fine-tuning method.
- What it does not show
- These are relative reductions on two test sets from a 2023 benchmark, not absolute error rates, and the recognizer was the one used in that paper. They do not describe a current dictation app.