Word Error Rate (WER)
Definition
The word error rate measures substitution, deletion, and insertion errors of automatic speech recognition.
An ASR metric: sum of substitutions, deletions, and insertions divided by the word count of the reference.
Simply explained
One compares the recognized transcript with the correct text and counts wrongly substituted, missing, and additional words.
Why it matters in practice
The term helps to select, develop, evaluate AI systems in a well-founded manner and realistically assess their technical risks.
Example from business
An operation measures WER separately for technical terms, noise, and speaker groups.
Limits and pitfalls
- Tokenization and normalization affect WER; the same rate can hide errors of different severity.
Origin & history
WER was established as a standard metric from edit distance and speech recognition evaluation.
Use in mid-sized companies
- Possible application: An operation measures WER separately for technical terms, noise, and speaker groups.
Frequently asked questions
What does word error rate (WER) mean?
An ASR metric: sum of substitutions, deletions, and insertions divided by the word count of the reference. What should be considered? Tokenization and normalization affect WER; the same rate can hide errors of different severity.
Sources
- SCTK scoring toolkit · NIST, SCTK scoring toolkit
Source checked on 10.10.2026. Editorial application examples are hypothetical; no evidence of real customer results.
Editorially reviewed on: