Glossary
Word Error Rate (WER) is the standard accuracy metric for speech recognition. It measures the percentage of words a transcription engine gets wrong by counting three error types — substitutions (wrong word), insertions (added word), and deletions (missed word) — divided by the total reference word count. Lower is better. A WER of 5% means 5 errors per 100 reference words.
How WER is calculated
WER = (S + I + D) / N, where S is substitution count, I is insertion count, D is deletion count, and N is the reference (correct) transcript word count. The math uses minimum-edit-distance alignment to count each error exactly once. WER can exceed 100% — if the model inserts more words than it gets right, you can have a 120% WER.
Real-world WER benchmarks (2026)
On the LibriSpeech clean test set (a standard benchmark of audiobooks), Whisper Large v3 achieves 1.8% WER. Real-world recordings are harder: 3-5% WER on clean podcast audio, 5-10% on conferenced calls, 10-20% on heavy accents or noisy environments. Human transcribers generally score 4-6% WER even when given the same audio.
Why WER alone is misleading
WER weighs every word equally — but missing the word "not" can flip the meaning of a sentence, while missing "the" doesn't. For real-world quality, complement WER with semantic-distance metrics, named-entity F1 (how many proper nouns are right), and human eyeball review on a sample. We optimize for the user-perceived quality more than the WER number.
In practice
Reference: "The CEO said we will not pursue the acquisition." Transcript: "The CEO said we will pursue the acquisition." That's a single substitution ("not" missed → 1 error / 9 words = 11% WER), but the meaning is now the opposite. WER of 11% sounds bad; the actual business consequence here is catastrophic.
Related terms
Further reading
See our accuracy in practice
Get started — freeLast updated: September 20, 2026
Frequently Asked Questions
What does WER mean in transcription?
Word Error Rate. The percentage of words a transcription engine gets wrong, including substitutions, insertions, and deletions. Lower is better. Modern engines achieve 2-5% WER on clean audio.
Why does WER matter when choosing a transcription workflow?
WER affects how you understand transcript quality, timing, compatibility, or the technology behind speech-to-text results.
Where can I apply WER on Transcript.you?
Use the related workflow at /ai-transcription when you want to see how this glossary concept connects to an actual transcription task.
Browse the full transcription glossary or read the complete guide to AI transcription.