Word error rate (WER)
Word error rate is the standard score for speech recognition: the share of words in a transcript that are wrong, missing or extra compared with a correct reference.
WER counts three kinds of mistake against a human-written reference: substituted words, deleted words and inserted words. It adds them up and divides by the number of words in the reference, so 5% means about one word in twenty is wrong. Lower is better, and the score can pass 100% if a model inserts many extra words.
Model cards report WER on public test sets, such as clean read speech, which are usually easier than real recordings. The same model can score very differently on meetings, phone calls, accents or noise, and the way text is normalized before scoring changes results, so numbers from different sources do not always line up. Test on a sample of your own audio. For Chinese and other languages written without spaces, character error rate is used instead.
Related terms
- Speech recognition (ASR, speech-to-text) Speech recognition, also called ASR or speech-to-text, turns spoken audio into written text.
- Benchmark A benchmark is a standardized test used to compare models, such as a set of maths, coding or knowledge questions; scores are useful but easy to over-read.