Overbark

Point your phone at a dog. Overbark reads what the moment sounds like it's saying — and floats the answer beside the dog as a comic speech bubble, stacking into a running chat that clears itself as the moment moves on. Underneath the deadpan is genuinely serious machinery, all of it running on the Apple Neural Engine. Each bark is isolated by Apple's on-device Sound Analysis classifier, then scored with Accelerate/vDSP spectral analysis across pitch, tonality, and inter-bark tempo. The Vision framework's animal body-pose tracking follows the dog's posture frame over frame — lying, sitting, standing, moving — while an on-device scene and sound classifier reads everything around the dog: weather, traffic, another dog barking, the occasional pigeon. A live observation deck — MINDSET, INSIGHTS, PRESENCE, VOCAL — reports it all, riding over a FFT soundwave drawn straight from the microphone. Point it at two dogs and each gets its own column of bubbles, replying to one another. Pick from three bubble voices — Editorial, Character, Comic — and eight harmonized accent themes, light or dark. An optional "Hooman Input Channel" lets your dog react to words you say too, on-device, shipped off by default. We stand by the sensors and technology, we do not stand by the sentence. Every line is a confident guess pointed at an animal that cannot confirm or deny it. 100% on-device. No account, no cloud, no tracking, no ads — and no cloud AI of any kind — not even Apple's own. The camera, the audio, and all the machine learning run right on your phone, in the moment; nothing is ever sent anywhere. Overbark never goes online at all — don't take our word for it, put your phone in Airplane Mode and it works exactly the same. Data collected: none. No subscription. No ads.