How does AI turn text into speech? Text-to-speech and voice cloning

AI reads text out loud

Last time, AI turned a voice into text. Now the other way around. The voice in this video is read by AI too. The machine does it in three steps.

STEP 1

Text into sounds

First, the machine turns text into the sounds it needs to make, called phonemes. The word "knight" has six letters, but only three sounds.

STEP 2

Add intonation

Then the machine works out the intonation: where the voice goes up, where it pauses, which word gets stressed. "You're home" can be a question or a statement, depending on intonation.

STEP 3

Make the sound wave

Finally, the model draws a spectrogram like in the last part, then turns it into a sound wave for the speaker. Because it learned from thousands of hours of real people reading, the voice even has breaths and feelings.

VOICE CLONING

A few seconds is enough to fake a voice

The worrying part: with just a few seconds of a voice sample, AI can copy someone's voice. Scammers can take your voice from videos you post, then call your family asking for money fast.

STAY ALERT

Strange calls asking for money

Remember the earlier video about fake faces? Now voices can be faked too. When a family member calls asking for money urgently, hang up and call back on the number you know.

PART 3

Listening and talking together

What happens when listening and talking come together? We chat with AI like with a real person. See you in part 3.