Fact · Speech recognition
17.6percent word error rate
On the CallHome test set of conversations between friends and family, Whisper Large V2 had a word error rate of 17.6 percent.
Checked against its source on October 17, 2026
In the source's own words
“Switchboard 28.3 13.8 51.2 CallHome 34.8 17.6 49.4”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Table 2, effective robustness across datasets.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- The table lists several systems per dataset. The value here is the Whisper Large V2 column, the same column as the Switchboard and CHiME6 figures.
- What it does not show
- One model on one test set. Read it beside the human baseline for the same kind of conversation.
Source
Cited on
- How to Transcribe Audio to TextGuide
- How Accurate Is Speech Recognition? The Real NumbersGuide
- Home Assistant Voice: Speech to Text and Voice CommandsGuide
- How to Transcribe an Interview: Audio to TextGuide
- Can you use AI transcription for research interviews?Question
- How do you transcribe a podcast?Question