Voice AI in ChatCubie – What's New?

Call quality in ChatCubie has had a round of work. Most of it is invisible when it is working, which is exactly the goal.

What changed

Noise suppression

Background noise — fans, traffic, a room with hard walls — is now filtered before your audio leaves the device. Processing on-device rather than on a server means it also works when the connection is poor, which is when you need it most.

Automatic gain control

Levels are evened out so someone sitting far from their microphone is not half as loud as someone speaking directly into theirs. In group calls this makes a bigger difference than any other single change.

Better behaviour on bad connections

When bandwidth drops, the call now degrades in a sensible order: video resolution first, then frame rate, and audio last. Audio is what carries the conversation, so it is what gets protected.

What we chose not to build

Two features came up repeatedly and were both declined:

• Automatic call transcription. It requires processing call audio somewhere we can read it. That undermines the privacy guarantee, and a private call is worth more than a searchable one.
• Live translation. Same problem, plus a second one — a mistranslation in a live conversation is worse than no translation, because both people believe they understood each other.

If a feature requires us to be able to read your conversations, the honest answer is not to build it.

On the roadmap

Echo cancellation when speakerphone is used in a small room, and better handling of the moment a call switches between wifi and mobile data — currently a brief gap, ideally imperceptible.

If calls sound worse for you after this update, that is genuinely useful to know. Tell us the device and network and we will look at it.