How do we know if a speech AI is good? We count its mistakes! There are three kinds: a wrong word, a missing word, and an extra word. Add them up and you get the Word Error Rate. Let's watch noise wreck it.
The AI is trying to transcribe a sentence it heard. Slide the noise up (like adding background music) and watch it make mistakes: wrong, missing, and extra words. The Word Error Rate counts them all.
background noise level
WER = (wrong + missing + extra) ÷ total words in the true sentence. A WER of 0% is perfect. 100% means as many mistakes as words! Real speech AIs are judged on this — lower is better.
Be the judge! Each card shows the true sentence and what the AI heard. What kind of mistake did it make?
You can now grade a speech AI like a scientist: count the wrong, missing, and extra words to get the Word Error Rate. Lower is better!