📊 Machines That Hear · Day 9

Scoring Speech AIs

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.

↓ count the errors
🎚️ Lab · Word Error Rate

Turn up the noise, watch it break

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.

✅ The true sentence:
🤖 What the AI heard:
0

background noise level

Word Error Rate
0%
wrong 0 · missing 0 · extra 0
🔍 Curious? The exact recipe

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.

🔎 Boss Build · Name That Error

Spot the mistake type

Be the judge! Each card shows the true sentence and what the AI heard. What kind of mistake did it make?

Correct: 0 / 4

🎉 Day 9 Complete!

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!

Day 10: Voice Cloning & Ethics →
Tomorrow: AI can now copy anyone's voice. Cool — but risky. Let's learn to spot fakes and be a responsible builder. 🛡️
AI Adventure · Machines That Hear · Day 9 — made for young inventors 🚀