Fact · Voice and AI
334,000pairs of N-best hypotheses and transcriptions
The HyPoradise benchmark contains more than 334,000 pairs of recognizer N-best hypotheses and their correct transcriptions, for training language models to correct speech recognition errors.
Checked against its source on October 17, 2026
In the source's own words
“encompassing more than 334,000 pairs of N-best hypotheses and corresponding accurate transcriptions”
HyPoradise: An Open Baseline for Generative Speech Recognition with Large Language Models, Chen Chen et al., arXiv (NeurIPS 2023 dataset track), 2023, Abstract.
Open the original- What it measures
- An N-best list is the recognizer's ranked list of alternative transcripts for one utterance. The dataset covers several speech domains and was built to test language-model error correction.
- What it does not show
- A research benchmark, not a product. The size shows how much training data the study used, not how accurate any app is.