Definition
Plain language
A score for how badly a transcript of speech gets the characters wrong.
As stated in the literature
Edit distance between a recognized transcript and the reference, normalized by reference length; because insertions are counted, it can exceed 100% when the recognizer hallucinates text absent from the audio.
Also called: CER
Why it matters: It gives a single number for how trustworthy a transcript is, and an unusually high score can reveal that the system is inventing text rather than merely mishearing it.
For example, if the reference sentence is "call mom" and the transcript says "call mob," one character out of eight is wrong, giving an error rate of about 13 percent.
Heard on the show
“On clean audio, the character error rate is around nine percent in English and fourteen percent in Chinese.”Episode 262 — Raise the Pitch Nine Percent and the Model Cries Sarcasm