Six microphones, one very quiet room

To hear you from across the room, Sonora needs to hear everything. The TV. The dishwasher. The conversation in the next room. And then ignore almost all of it.

That problem has a name in audio engineering: far-field voice. It's the difference between a device you have to lean into and one you can talk to from the couch. Getting it right took six microphones, a lot of math, and one extremely quiet room.

Why six

One microphone hears a blur. It can't tell where a sound came from, so it can't separate your voice from the noise around it.

With six microphones arranged around the device, Sonora can do something closer to what your own ears do:

• Figure out which direction your voice is coming from

• Focus on that direction and turn down everything else

• Track you as you move around the room

• Cancel out steady background noise like a fan or AC

This is called beamforming. The microphones work as a team, and the on-device chip does the heavy lifting in real time, before a single word is sent anywhere.

The quiet room

To tune all of this, we built in an anechoic chamber, a room engineered to have almost no echo at all. The walls, floor, and ceiling are covered in foam wedges that absorb sound instead of reflecting it. Stand inside and your own voice sounds strangely flat, like the room is swallowing it.

We needed that silence to measure precisely. In a normal room, sound bounces off every surface and muddies the data. In the chamber, we could test exactly how Sonora hears a voice from three feet, ten feet, at an angle, with a fan running, with music playing.

We ran thousands of these tests. A few things we learned:

• Hard surfaces nearby, like a tile backsplash, change how voice reaches the device. Sonora adapts to its surroundings in the first few seconds after setup.

• The hardest sound to talk over isn't loud music. It's another person talking. Human voices live in the same frequency range, so separating them is the real challenge.

• Distance matters less than we expected. Angle matters more.

Doing it on-device

The easy way to handle all this processing is to send audio to the cloud and let a big server figure it out. We didn't want to do that. Sending your voice out of the room is exactly what we built Sonora to avoid.

So the beamforming, the noise cancellation, the wake-word detection, all of it runs on the chip inside the device. Your voice is processed where you're standing and never stored. That constraint made the engineering harder. It also made Sonora the device we actually wanted to put in our own homes.

What it feels like

All of this disappears in the final product, which is the point. You don't think about six microphones or beamforming or anechoic chambers. You just talk, from wherever you are, and Sonora hears you.

That's the whole goal. The hard part is ours. The easy part is yours.