Course 2, lesson 17 of 100, Ages 6+
How AI listens
Turning sound into words
Like I’m 5
Your voice makes wiggly waves in the air, like ripples in a pond. AI learns which wiggles make which words.
The big idea
Sound is a wiggly wave travelling through the air. A microphone measures that wave thousands of times every second, turning your voice into a long list of numbers.
Speech AI has heard millions of voices, so it learned which wave shapes match which sounds and words. It also uses the words around them to guess better, so 'ice cream' doesn't become 'I scream'.
Examples
- Subtitles: Videos can make captions automatically from speech.
- Voice typing: Speak a message and it appears as text.
- Accents: AI understands more accents when it was trained on many kinds of voices.
How it works
- Sound is wiggly waves in the air. A microphone turns those waves into numbers.
- AI learns which wiggles match which sounds, and which sounds make which words.
- That’s how captions appear on videos and how phones type what you say.
Check your understanding
- How does AI hear a sound?
- Options: As numbers from a microphone; Through tiny ears; It can’t hear anything.
Answer: As numbers from a microphone. The microphone measures the sound wave and turns it into numbers the AI can learn from. - Why does speech AI need voices from many people?
- Options: So it understands lots of accents and ages; So it can sing; So it gets louder.
Answer: So it understands lots of accents and ages. Variety in voices helps the AI understand everyone.
Remember
Microphones turn sound into numbers that AI can learn from.
Talk about it
Which sounds are hardest for you to understand?
Go deeper
Speech recognition turns audio into spectrograms, pictures of sound over time, and predicts the most likely words. Systems learn from huge amounts of recorded speech, and still struggle more with accents and dialects that are rare in their training data.