Fact · Transcription
16.9percent word error rate
On the AMI meeting corpus recorded with individual headset microphones (IHM), Whisper Large V2 had a word error rate of 16.9 percent.
Checked against its source on October 17, 2026
In the source's own words
“CORAAL 35.6 16.2 54.5 AMI IHM 37.0 16.9 54.3 Switchboard 28.3 13.8 51.2”
Robust Speech Recognition via Large-Scale Weak Supervision, Alec Radford et al., arXiv (eess.AS), 2022, Table 2, effective robustness across datasets, AMI IHM row.
Open the original- What it measures
- AMI is a corpus of real multi-person meetings. In the IHM condition each speaker wears a close headset microphone. In the paper's table the middle column is Whisper Large V2, and the first column is the supervised comparison model that the paper sets beside it.
- What it does not show
- Results depend on the normalizer the paper used to score transcripts. The meetings are research recordings, not a particular meeting app. This is a 2022 model, and newer models may score differently.