Word error rate counts substitutions, insertions, and deletions against a reference transcript, divided by the number of words. A 5% WER means one word in twenty is off. It is the yardstick for choosing and tuning ASR, though it ignores which errors hurt most.
The book · Volume 6 of 7
AI for Video: A Complete Guide to Intelligence in Video Products: From Models and Inference to Computer Vision, Speech, Search, and Generative Video
WER is the number every ASR vendor quotes, yet it is only meaningful on your own audio and with consistent text normalisation. Volume 6 covers WER in detail within speech AI: measuring ASR accuracy, comparing engines, diarization, TTS and how transcript quality affects search and downstream features. Written by Nikolay Sapunov, CEO at Fora Soft.
Need a fair WER comparison of ASR engines on your own audio?
Send us a sample of your recordings, and we will benchmark ASR engines by WER on them and recommend one with cost per hour. 250+ video products shipped since 2005.
