Fact · Dictation software
The study defines word error rate with the total number of words in the correct transcript as its denominator.
Checked against its source on October 16, 2026
In the source's own words
“is the total number of words in the ground truth”
Racial disparities in automated speech recognition, Allison Koenecke et al., Proceedings of the National Academy of Sciences, 2020, Methods, Word error rate (text after Equation 1).
Open the original (DOI 10.1073/pnas.1915768117)- What it measures
- The study compares five commercial speech recognition systems against human transcripts using this measure.
- What it does not show
- The denominator is the study's definition. The measure counts every word error the same way, so it does not measure how much a mistake costs in editing time.
Source
Cited on
- Speech Recognition: What It Is and How Well It WorksTopic
- Transcription: How to Turn Audio and Video Into TextTopic
- How Speech Recognition Works, From Sound to TextGuide
- Best Dictation Software and Apps, Compared by NeedGuide
- Word Error Rate (WER): How to Calculate and Read ItGuide
- How Accurate Is Speech Recognition? The Real NumbersGuide
- Speech Recognition and Accents: Accuracy and FixesGuide
- Transcription Software: How to Choose OneGuide
- How do you calculate word error rate?Question
- Why does speech recognition get names and jargon wrong?Question
- Word error rate (WER)Glossary