WER vs CER: both scores count substitutions

Both scores count substitutions, deletions and insertions against a reference and divide by its length. WER does it over words, CER over characters.

Both scores count substitutions, deletions and insertions against a reference and divide by its length. WER does it over words, CER over characters. A near-miss spelling is a whole wrong word under WER but only a few wrong characters under CER, so CER separates small slips from wrong words and is kinder to languages with marks and special letters.

Also searched as "cer or wer" and "difference between wer and cer".

Character error rate and word error rate side by side

TermWhat it isMain source
Character error rateCharacter error rate is word error rate computed on characters instead of words: substitutions, deletions and insertions of characters divided by the number of characters in the reference.lightning.ai, char error rate
Word error rateWord error rate is the standard score for a speech recognition transcript.Wikipedia, Word error rate

What the sources say about WER vs CER

How Autrans treats the difference

Autrans' published per-language figures are word-based, so a missing tone mark costs a whole word. That is why restoring marks for Yoruba and Igbo matters to its scores, and why a reader should compare a figure only with figures measured the same way on the same kind of audio.

A worked example

Reference: 'Ọlá ra ọkọ tuntun' (4 words, 14 characters ignoring spaces). Transcript: 'Ola ra oko tuntun'. Two words wrong, so the word score is 2 divided by 4, 50 in 100. Four characters wrong (Ọ, á, ọ, ọ): the character score is 4 divided by 14, about 29 in 100.

Related terms

Questions people ask

What is the difference between WER and CER?

Both scores count substitutions, deletions and insertions against a reference and divide by its length. WER does it over words, CER over characters.

Does the difference change anything in Autrans?

Autrans' published per-language figures are word-based, so a missing tone mark costs a whole word. That is why restoring marks for Yoruba and Igbo matters to its scores, and why a reader should compare a figure only with figures measured the same way on the same kind of audio.

Where does this definition of WER vs CER come from?

It rests on Wikipedia, Word error rate and lightning.ai, char error rate. What each source supports is listed above, next to its link.

Can you give an example?

Reference: 'Ọlá ra ọkọ tuntun' (4 words, 14 characters ignoring spaces). Transcript: 'Ola ra oko tuntun'. Two words wrong, so the word score is 2 divided by 4, 50 in 100. Four characters wrong (Ọ, á, ọ, ọ): the character score is 4 divided by 14, about 29 in 100.

Related

Start transcribing free

Send Autrans into your next meeting, use it in the room, or upload a recording. Your first 30 minutes are free, given once, and no credit card is needed to start.

Last checked 22 October 2026. How we measure accuracy. Spotted a mistake? Write to hello@autrans.online.