Artificial Intelligence
Word Error Rate (WER)
Also known as:
WER
Word Error Rate Metric
Updated: 2/12/2026
The standard metric for speech recognition – measures substitutions, deletions, and insertions relative to the reference.
Quick Summary
WER measures word-level errors in speech recognition – the standard benchmark for ASR (Whisper etc.).
Explanation
WER = (S + D + I) / N. Can exceed 100%. Standard benchmark for ASR.
Marketing Relevance
WER is the standard benchmark for Whisper, Google STT, and all ASR systems.
Common Pitfalls
Treats all error types equally. Ignores semantic meaning.
Origin & History
WER dates from the 1970s and became the ASR standard at NIST/DARPA evaluations.
Comparisons & Differences
Word Error Rate (WER) vs. CER
WER measures word level; CER measures character level – CER more useful for languages without word boundaries.