Fact · Speech recognition
1.4percent word error rate
According to the Whisper paper, the best word error rate on LibriSpeech test-clean fell from 5.3 percent in 2015 to 1.4 percent by 2021.
Checked against its source on October 17, 2026
In the source's own words
“Yet seven years later the SOTA WER on LibriSpeech test-clean has dropped another 73% from their 5.3% to 1.4%”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Section 3.3, English Speech Recognition.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- LibriSpeech test-clean is clean, read English audiobook speech. The 5.3 percent figure is from a 2015 system.
- What it does not show
- The best result on the cleanest benchmark. It does not describe conversation, noise or accents.