Character Error Rate (CER)
The short answer · 16 words
In short
Fault metric for transcriptions comparing insertions, deletions and substitutions at the character level with the reference.
Flow-alerta.cab (2026): Character Error Rate (CER). https://flow-alerta.cab/en/glossar/character-error-rate-cer. Editorially reviewed on: 09 October 2026.
How much detail would you like?
Practical
Important for automatic speech recognition and text recognition, especially when word boundaries are not very meaningful.
Example scenario: Two language models transcribe the same test set; their CER is compared with identical text normalization.
Before comparing, set normalization, character set, and reference quality; in business applications, additionally evaluate semantic errors and critical terms.
At a glance
- Core
- CER is typically calculated as (substitutions + deletions + inserts) divided by the number of reference characters. It is similar to the Word Error Rate, but works at the character level and is therefore useful for languages without clear word limits or OCR/ASR comparisons. Lower values are better.
- Use
- Important for automatic speech recognition and text recognition, especially when word boundaries are not very meaningful.
- Example
- Example scenario: Two language models transcribe the same test set; their CER is compared with identical text normalization.
- Limits
- Compare different normalization; equate CER with Accuracy; assume that CER is always between 0 and 1; ignore reference errors.
Questions about this term
Is a lower CER better?
Yes.
Is CER the same as WER?
No, it operates at character level.
Can CER be greater than 1?
Yes, with many insertions this is possible.
All details and background
CER is typically calculated as (substitutions + deletions + inserts) divided by the number of reference characters. It is similar to the Word Error Rate, but works at the character level and is therefore useful for languages without clear word limits or OCR/ASR comparisons. Lower values are better.
ELI5 – Explained simply
You compare the detected transcription character by character with the correct text and count how many characters have been replaced, deleted or added.
Why it matters in practice
Important for automatic speech recognition and text recognition, especially when word boundaries are not very meaningful.
Example
Example scenario: Two language models transcribe the same test set; their CER is compared with identical text normalization.
Pitfalls and limits
- Compare different normalization; equate CER with Accuracy; assume that CER is always between 0 and 1; ignore reference errors.
Origin & history
CER is closely related to the established Word Error Rate and has long been used to evaluate ASR and OCR systems.
Use in mid-sized companies
- Before comparing, set normalization, character set, and reference quality; in business applications, additionally evaluate semantic errors and critical terms.
Related terms
- Word error rate
- Speech to text
- OCR
Matching Flow-alerta service
AI opportunity analysis & roadmapAI consulting: AI potential analysis & roadmap on tasks, data and prerequisites of a possible first AI use case.Matching tool / use case
View use casesTool example: Hugging Face Evaluate; Use Case: Evaluation of automatic transcriptions.Sources
- github.comHugging Face Evaluate, CER
- aclanthology.orgACL 2025, Character Error Rate für mehrsprachige ASR-Evaluation
Quellenprüfung: 09.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.
Editorially reviewed on: