Work out character error rate instantly with clear inputs, formula shown and shareable results.
Character error rate applies the same edit-distance ratio as WER but at character granularity, which makes it the standard metric for OCR, handwriting recognition and languages without clear word boundaries such as Chinese and Japanese. CER is always lower than WER on the same output because a single wrong letter costs one character error instead of a whole word.
Character error rate
CER = (S + D + I) / N where N is the number of reference characters
That is a convention you must fix and document. Including whitespace is common for OCR; excluding it makes CER slightly lower and is usual for scripts without spaces.
Scanned text often has single-glyph confusions such as rn for m. WER marks the whole word wrong, hiding the fact that recognition was 90 percent correct at the glyph level.