Fact · Speech recognition
2.9percent average word error rate
In the same 2024 study, Whisper large-v2 had an average word error rate of 2.9 percent on the English lecture recordings.
Checked against its source on October 17, 2026
In the source's own words
“Speechmatics 3.3 3.6 4.6 8.0 Tencent 4.6 4.1 7.3 - Whisper (large-v2) 2.9 3.3 3.3 5.0”
Measuring the Accuracy of Automatic Speech Recognition Solutions, Korbinian Kuhn et al., arXiv (preprint), 2024, Table 4.
Open the original- What it measures
- The table lists Whisper (large-v2) alongside the commercial services, on the same recordings and settings.
- What it does not show
- Same study and same caveats as the Google figure. A difference of this size in one test does not settle which service is better for your audio.
Source
Used to calculate
Cited on
- Zoom, Teams and Google Meet TranscriptionGuide
- Video to Text: Transcribe Videos and YouTubeGuide
- Transcription Software: How to Choose OneGuide
- Speech Recognition in Python: Libraries ComparedGuide
- How do you transcribe a lecture recording?Question
- How do you transcribe a Zoom audio or video recording?Question
- Is Whisper more accurate than Google Speech-to-Text?Question
- What is IBM Watson Speech to Text?Question