Traditional walkie-talkies are bulky and clumsy, and you have to press a button to talk.

One day while riding, I wondered — could I make a portable, hands-free walkie-talkie?

So I DIYed a walkie-talkie that’s a bit different. It uses an embedded neural network ? to automatically detect human voice ?️, recognize keywords to trigger animations, and even turn other conversations into text bubbles. It’s a lot of fun.

A picture is worth a thousand words.

Here is the video (English Closed Captions available):

When we think of walkie-talkies, they’re usually bulky, clunky, and limited to basic voice communication. During a bike ride one day, I realized how difficult it was to communicate with friends riding behind me.

That got me thinking — could I build a compact, hands-free walkie-talkie that doesn’t require pressing any buttons?

Traditional walkie-talkies use half-duplex communication, meaning only one person can speak at a time. You have to hold down the PTT button to talk, but when you’re riding a bike, that’s simply not practical.

Traditional walkie-talkies use half-duplex communication, meaning only one person can speak at a time. You have to hold down the PTT button to talk, but when you’re riding a bike, that’s simply not practical.

So I used an AI model (VADNet – Voice Activity Detection Neural Network) to automatically detect human speech and switch between transmit and receive modes. This completely eliminates the need for a push-to-talk button.

I also added a high-brightness grayscale OLED display to show system status. With speech recognition, triggering specific keywords causes corresponding animations to appear on the screen, making the device more interactive and fun. Even speech that doesn’t match any keywords can still be converted into text and displayed as chat-style speech bubbles.