What the Neural Engine inside M1 through M5 chips does, which Whisper runtimes actually use it, and how to measure local transcription speed on your own Mac.
Word Error Rate is the standard metric for transcription accuracy, but most dictation users have never calculated it. Here's how to benchmark your setup and fix what's wrong.
How multilingual speech models detect language, why some language pairs are harder than others, and the dictation habits that help bilingual professionals switch smoothly.
A plain-language breakdown of the Neural Engine, GPU, and unified memory, and how they let a speech model like Whisper transcribe on your Mac without a network connection.
How Whisper model sizes differ in parameters, memory, and relative speed, what OpenAI publishes about each, and how to benchmark the right model for your own dictation.