What is character error rate?

By the Potok teamUpdated ~1 min read

What is character error rate? It is the share of characters that a system gets wrong compared with a reference text, counted as substitutions, deletions and insertions. CER is used mainly for languages written without spaces, where splitting text into words is unreliable.

Transcribe a recording

Send your voice message, audio or video to @potok_now_bot on Telegram and the text comes back in the chat.

Open @potok_now_bot in Telegram

Files up to 2 GB · export to TXT, MD, DOCX, PDF, SRT and VTT · speaker labels and timestamps · about 30 languages

Key facts

  • The formula is the sum of substituted, deleted and inserted characters, divided by the number of characters in the reference.
  • Word error rate treats one wrong letter as a whole wrong word, so CER gives a gentler view of small spelling slips.
  • Japanese and Chinese text has no spaces between words, so a word count depends on a segmentation rule; characters do not.
  • A CER of zero means an exact match with the reference, and the value can exceed one when many extra characters appear.

What it looks like

CER = (S + D + I) / N
  S substituted, D deleted, I inserted characters; N = characters in reference

reference : recognise    (N = 9)
output    : recognize
S = 1 (s -> z), D = 0, I = 0
CER = 1 / 9 = 0.11

More

See the full list of languages, the tools and the glossary.

Frequently asked questions

What is the difference between CER and WER?

WER counts errors per word and CER per character. CER is finer, and it suits scripts where word boundaries are not marked by spaces.

How do I calculate CER?

Align the output with the reference, count substituted, deleted and inserted characters, and divide the total by the reference length in characters.

Why not use WER for every language?

WER needs a clear definition of a word. In languages written without spaces that definition varies by tool, so scores from different tools stop being comparable.

Is a lower CER always better?

Yes for the same test text. Lower means fewer character mistakes, but compare only scores measured on the same reference, with the same text rules.

Do published CER numbers apply to my file?

Not directly. They come from specific test sets, and a noisy or accented recording will score differently. Treat them as a guide, then check your own file.

Sources

Each source was opened by hand on the date shown. Pages change, so open the source if a detail matters to you.

  1. Word error rate (Wikipedia), Wikipedia · checked
  2. Japanese writing system (Wikipedia), Wikipedia · checked

Try it on your own recording

Send your voice message, audio or video to @potok_now_bot on Telegram and the text comes back in the chat.

Open @potok_now_bot in Telegram