Apple's iOS 16 dictation runs fully on-device, and that single design choice has cascading consequences for latency, privacy, and what kinds of ASR architectures are even viable on a phone. A speech researcher breaks down what Apple is likely doing under the hood and why the A15 Bionic's neural engine is finally powerful enough to skip the round trip to a server.

The walkthrough gets into practical questions engineers actually care about: how the A15's ARM-based silicon stacks up against Intel Xeon parts for neural inference, whether you can realistically deploy your own custom ASR models on iOS, and what constraints on-device speech recognition places on model size, quantization, and streaming decoding. It's a useful primer for anyone considering an edge-first speech stack or curious about how the major platform vendors are quietly redrawing the client-server boundary for voice. Give it a look if you're weighing on-device versus cloud ASR for a product decision.