SoundSleep
A sound classifier is trained on the world in general. Your bedroom is not the world in general. It has one specific fan, one specific partner, a radiator that knocks, and a neighbour who comes home at midnight.
A general model will call that fan an “appliance” forever, which is technically right and personally useless. So we stopped trying to be right in general and started letting you be right in particular.
Your words, saved
When SoundSleep catches something it could not place, it asks what it was. Alongside the usual answers you can type your own: my fan, the dog, my partner, radiator. That label is saved. Next time it is a chip you tap once, and the labels you use most float to the front.
The same applies to the Room Report. Tap any sound it identified and you can rename it, label individual moments, or hide it entirely — after which it stops being reported on future nights. If the fan is always on and you do not care, the app should stop telling you about the fan.
Why this is not busywork
Two reasons. The first is that a report in your own vocabulary is a report you will read. “My fan, six times” lands differently from “mechanical appliance, six detections.”
The second is that labelling is the only thing that can separate a sound that matters from one that does not, and no amount of model tuning replaces a person who was in the room. Every label you give makes the next night's report sharper.
So we made it a game
Naming sounds earns Sound Trainer XP, which counts toward your level, and there are ranks along the way from First Label to Sound Sage, plus five badges for the people who really commit. It is a small thing. It also turns a chore into something people actually do.
It stays on your phone
Your labels are yours. They are stored on the device with everything else and are never uploaded, pooled, or used to train anything. The app learns your room; nobody learns about you.
← All posts