Expired60% confidenceBenchmarkExact time
On the LibriSpeech test-clean benchmark, Whisper large-v3's WER has dropped to around 2%, approaching or in some cases surpassing human annotators (approximately 4–5%).
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/13/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 10/11/2026(expired)
Sources
Why Does Speech-to-Text Keep Failing? A Practical Guide to Optimizing Messy Audio Transcription
redditr/LanguageTechnology7/11/2026
Related Claims
UnverifiedWhisper has been iterated to the large-v3 version (November 2023), with an average word error rate (WER) on the Fleurs benchmark reduced to approximately below 10%.78% similarUnverifiedOn the LibriSpeech benchmark, Oído's word error rate (WER) is 3.7 (clean) and 8.2 (other), while Whisper tiny.en on a laptop achieves 6.3 and 15.9.69% similarUnverifiedOn the SWE-Bench Multilingual benchmark, HY3 scored 75.8, slightly behind DeepSeek V4 Pro's 76.2.60% similarUnverifiedWhisper large-v3 是目前精度最高的版本，约15亿参数，采用 MIT License60% similarUnverifiedOn the test-other set, B-WER was reduced by 23.9% relative to the CTC-Filter baseline.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/497314API
curl https://kongchang.com/api/v1/knowledge/claims/497314MCP
get_claim(id=497314)