Fact · Transcription
36.4percent word error rate
On the AMI meeting corpus recorded with a single distant microphone (SDM1), Whisper Large V2 had a word error rate of 36.4 percent.
Checked against its source on October 17, 2026
In the source's own words
“AMI SDM1 67.6 36.4 46.2 LibriSpeech Other 6.2 5.2 16.1”
Robust Speech Recognition via Large-Scale Weak Supervision, Alec Radford et al., arXiv (eess.AS), 2022, Table 2, effective robustness across datasets, AMI SDM1 row.
Open the original- What it measures
- The SDM1 condition is one microphone placed away from the speakers, as in a room or a laptop on a table. In the paper's table the middle column is Whisper Large V2, the same model as in the headset result.
- What it does not show
- One microphone in one set of meeting rooms is not a measure of every laptop or room. The result shows the direction of the effect for the same model, not a guaranteed error rate for your meetings.