Talking with your voice
Last time, AI read text out loud. Now we put listening and talking together. You ask out loud, and AI answers out loud, like calling a friend. But inside, there are two very different ways to do it.
THE OLD WAY
Three machines in a row
The old way chains three machines: one transcribes your words into text, AI writes a reply, and a third machine reads it out loud. Each stage takes a moment, so you wait a few seconds in total.
THE OLD WAY
Your tone gets lost
Worse, once it becomes text, your tone of voice disappears. You say "yeah, fine" in a sad voice, AI only sees those two words, and answers cheerfully.
THE NEW WAY
One brain that listens and talks
The new way uses one model: it hears the sound directly and speaks sound directly, without turning it into text. It can notice a sad voice or laughter, and answer in a tone that fits you.
FAST
Answers in about half a second
With fewer stages, AI can answer in about half a second, close to the pace of human conversation. You can even interrupt: if you cut in, it stops to listen.
NOTE
A warm voice isn't always right
But a natural-sounding voice doesn't mean the information is right. For important things like medicine or money, still double-check.
PART 4
Next up: watching video
Listening and talking, done. Now on to watching a whole video. How does AI watch video? See you in part 4.
This article is based on the video Talking with your voice from the Mark học AI channel. Watch the video (in Vietnamese) to see the animations.