WER is typically measured on standard test sets with clear pronunciation and quiet backgrounds, and real-world performance in actual meeting environments is often worse than official figures.