Word Error Rate (WER) & CER Calculator
Speech-recognition and OCR accuracy — substitutions, deletions and insertions per reference word/character.
WER is Levenshtein distance computed over words, divided by reference length. It can exceed 100% (insertions aren't capped). Human transcribers sit around 4–5% WER on clean speech; modern ASR matches this on benchmarks but degrades with accents, noise and jargon. CER uses characters — better for tonal/agglutinative languages.
Formula
About Word Error Rate (WER) & CER Calculator
Word Error Rate is the universal yardstick for speech recognition and machine transcription: it counts the substitutions, deletions and insertions needed to turn the model's output into the reference transcript, divided by the number of reference words. It's literally Levenshtein edit distance applied to word tokens. This calculator computes WER (or character-level CER for languages where word boundaries are fuzzy), the raw edit count, and accuracy. Human transcribers achieve ~4–5% WER on clean audio; this tool lets you measure how close your ASR or OCR system gets.
How to use Word Error Rate (WER) & CER Calculator
- 1Enter your values into Word Error Rate (WER) & CER Calculator — sensible, domain-typical defaults are pre-filled so you see a real result immediately.
- 2The result recomputes live using the formula shown on the page; there is no button to press.
- 3Adjust any input to compare scenarios, then read the worked example to see the substituted numbers.
Why use Word Error Rate (WER) & CER Calculator?
- ✓Computes Word Error Rate (WER) & CER instantly in your browser — no sign-up, no upload, no server round-trip.
- ✓100% free and unlimited, with the exact formula shown: WER = (Substitutions + Deletions + Insertions) / reference words — Levenshtein distance over word tokens, normalized.
- ✓Runs entirely client-side, so every value you enter stays private on your device.
- ✓Live recompute as you type, with a worked example and authoritative references for trust.
Frequently asked questions
Can WER be greater than 100%?+
Yes — because insertions aren't bounded by the reference length. If the model hallucinates many extra words, the insertion count can exceed the reference word count, pushing WER above 100%. This is correct behavior, not a bug; it flags a model that's adding far more than it's getting right.
WER or CER — which should I use?+
WER for languages with clear word boundaries and where word-level errors matter (English, most European languages). CER (character error rate) for languages without spaces (Chinese, Japanese), agglutinative or highly inflected languages, and OCR, where word segmentation is ambiguous or a single character error shouldn't fail a whole long word. Many Asian-language ASR systems report CER.
Why doesn't WER capture meaning?+
It's purely lexical — 'can't' vs 'cannot' is an error, and a semantically catastrophic substitution ('can' → 'can't') counts the same as a trivial one. It also ignores punctuation and capitalization unless you include them. For meaning-aware evaluation, pair WER with semantic or downstream-task metrics; WER measures transcription fidelity, not comprehension.
How do I reduce WER on my domain?+
Common levers: fine-tune or adapt the acoustic/language model on in-domain audio and vocabulary, add a domain language model or biasing list for jargon and names, improve audio quality (denoising, better mics), and handle accents with diverse training data. Measure WER on a held-out set representative of your real conditions, not just clean benchmark speech.
Related ML & AI tools
ROC-AUC Calculator (from TPR/FPR points)
Trapezoidal area under the ROC curve from your (FPR, TPR) operating points — the threshold-independent ranking score.
● LiveClassification Threshold Cost Calculator
Find the probability cutoff that minimizes expected cost given your false-positive and false-negative penalties.
● LiveSilhouette Score Calculator
Cluster cohesion vs separation for one point — the building block of the silhouette metric for choosing K.
● Live