Press Kit

Longer

Annotate Audio teaches an iPhone to recognise the sounds its owner cares about: a dog barking, a snore, a cough, a baby crying, a doorbell. Each sound gets its own folder, and every recording in it is a labelled example. Five recordings each of two sounds are enough to train a first model, and a “Not …” folder with the other sounds of the room teaches it the difference.

Training happens on the iPhone itself, with Apple's Create ML, in a few minutes and without writing code. The finished Core ML model then listens live and records a clip every time it hears one of the trained sounds, filed into that sound's folder. A rolling pre-buffer means every clip starts before the sound did, so it can listen overnight or while nobody is home without leaving hours of silence to scroll through.

A pixel pup trains alongside: every recording is a rep, and it levels up from Puppy to Learning, Good dog and Top dog, earning a collar at 5 reps, a tag at 50 and a crown at 100. For makers and researchers, the app doubles as a quick way to collect and label audio datasets in the field and export them, laid out for Create ML on a Mac.