Fact · Speech recognition
5.1percent word error rate
In the IBM study, the best of three human transcribers reached a word error rate of 5.1 percent on the Switchboard telephone test after quality checks.
Checked against its source on October 17, 2026
In the source's own words
“could only get down to 5.1% for the best transcriber after quality checks”
English Conversational Telephone Speech Recognition by Humans and Machines, George Saon et al., arXiv (cs.CL), IBM Research, 2017, Section 2, human transcription experiments.
Open the original- What it measures
- Three transcribers from the vendor Appen each transcribed the same calls, and a fourth transcriber checked their work.
- What it does not show
- Telephone conversations, not dictation into a microphone. The number is a benchmark for conversational speech, not for dictation into a microphone.