Definition
Word error rate is Word error rate (WER) is the standard score for speech-to-text accuracy: the share of words a system gets wrong compared with a correct transcript, counting substitutions, deletions and insertions, so lower is better.
Word error rate, or WER, is the most common way to score a speech recognition system. You take a human-checked "reference" transcript, compare it with what the system produced, and count the mistakes. A WER of 5% means roughly five errors per hundred reference words. Lower is better.
WER counts three kinds of error:
The score is (S + D + I) divided by N, where N is the number of words in the reference. It is built on word-level edit distance, the minimum number of changes needed to turn one text into the other.
A quick example: if the reference is "please turn around" and the system writes "please burn a round," that is one substitution, one deletion and two insertions across three reference words, for a WER above 100%. Because insertions count, WER can exceed 100%.
WER is useful and blunt. Every error weighs the same, so a dropped "the" costs as much as a mistaken drug name. It also depends heavily on the test set: clean studio audio, accents, noise, overlapping speakers and specialized vocabulary all move the number. Two vendors quoting different WERs may simply have tested on different audio.
WER belongs to the family of AI benchmarks: a fixed test with a scoring rule. Transcripts feed voice agents, and tools such as evals let you measure accuracy on your own recordings. The opposite direction, turning text into audio, is text-to-speech.
It depends on the audio. Clean, single-speaker recordings can reach low single digits, while noisy calls or heavy accents score far higher. Compare systems only on the same test audio.
Yes. Insertions count as errors, so a system that adds many extra words can make errors outnumber the words in the reference.
Mistral released Voxtral 3 on Oct. 20: new transcription and speech models plus a voice agent API at $0.04 a minute when it reasons with Mistral Small 4. Skip the benchmark charts. The per-minute price and the choice of where audio is processed are the story.