Blog
How Advanced AI noise cancellation works
Published October 29, 2026
Every call has the same problem: your microphone hears everything. The keyboard you type notes on, the fan, the dog next door, a truck going by. Huddles and meetings in Freesbe have three answers to that, under Preferences → Calls & audio → Audio: Off, Standard and Advanced AI. This post explains what each one does, how the strongest one works, and why we made it the way we did.
Off, Standard and Advanced AI
Off sends your microphone as it is. That’s right for music, or a good studio microphone in a quiet room, where any filtering takes away more than it removes.
Standard is the noise filter built into your browser and the desktop app. It’s classic signal processing: it listens for sound that stays the same over time, like a fan or an air conditioner, and turns it down. It costs almost nothing and it’s good at steady hum. It isn’t good at sudden noises, because a key press or a bark looks a lot like the start of a word.
Advanced AI tells your voice apart from everything else, including the sudden noises. That’s the one people mean when they say “AI noise cancellation”, and it’s the rest of this post.
How Advanced AI works
Advanced AI is built on RNNoise, an open-source noise suppressor from the people behind the Opus audio codec that calls already use. It is a small neural network: a model trained on many hours of recordings of speech mixed with noise, until it learned to tell which is which.
A hundred times a second, the app takes the last 10 milliseconds of sound from your microphone and splits it into frequency bands, from the low hum of a fan to the hiss of an “s”. The network looks at those bands, and at what it heard just before, and decides for each band how much of it is voice. Bands that are mostly voice pass through. Bands that are mostly keyboard, dog or traffic are turned down. Because the network remembers the last moments of sound, it can tell the click of a key from the start of a word, which is exactly where the simple filter gets it wrong.
The whole filter, model included, is about 150 kilobytes, so it runs on your own computer, in the same place the app already processes your microphone. On a typical laptop it uses a small share of one processor core.
Is this active noise cancellation?
Not quite, and the two work well together. Active noise cancellation (ANC) is what noise-cancelling headphones and earbuds do: tiny microphones on the outside of each ear cup hear the noise around you, and the speakers play the exact opposite sound wave, so the two cancel out before they reach your ears. ANC protects what you hear.
Advanced AI protects what the others hear. It cleans up your microphone before your voice leaves your computer, so the people on the call don’t hear your keyboard, your dog or the street. Your headphones can’t do that part: even the best ANC headphones send your microphone with the room in it.
So if you have ANC headphones, keep them on for your own ears and turn on Advanced AI for everyone else’s. Together you get a quiet call in both directions.
Why on your computer
We could have sent your audio to a server to clean it up. We didn’t, for three reasons:
- Privacy. Your raw microphone never goes anywhere it wasn’t already going. The filter runs before your sound leaves your computer, and nothing about what you say is stored or sent anywhere extra.
- No delay. A trip to a server and back would add a delay to every word. On your computer, the filter adds about 10 milliseconds, which nobody can hear.
- It works everywhere. It doesn’t matter which call server the call runs on, or whether the others use the browser, the desktop app or a phone.
Why it isn’t the default
Standard stays the default for now. Advanced AI uses a little more battery, which matters on an older laptop running a long meeting. It also treats anything that isn’t speech as noise, so if you play music to the others, or want them to hear the room, Off or Standard is better. You can switch at any time, even in the middle of a call, from Audio and video devices.
Making sure you’re still heard
A filter that decides what to send has one way to fail that matters more than any other: sending silence while you talk. So every call checks for it. Every few seconds, Freesbe compares how loud your microphone is with what the call actually sends. If you’re talking and the others are getting silence, the call switches to Standard by itself within about 10 seconds and tells you. Your setting stays as you chose it for the next call.
To try it, open Preferences → Calls & audio → Audio, choose Advanced AI and click Test microphone, then type on your keyboard while you talk: the bar follows your voice and stays down for the keys. See Noise cancellation in the help center for the details.