Core ML
Apple's on-device model runtime. It decides whether a model runs on the Neural Engine, the GPU, or the CPU.
Core ML is the framework that runs machine-learning models on Apple hardware. A model is converted into Apple's .mlpackage / .mlmodelc format ahead of time, and at load time Core ML compiles it for the machine it is running on and spreads the work across the Neural Engine, the GPU, and the CPU.
How that work is spread is not fixed: it depends on the chip and the macOS version, and the difference between a well-matched setup and a poor one can be several times the speed on identical audio. It is the main reason transcription benchmarks from one Mac rarely reproduce on another. Mac Note Taker tunes this for your Mac on first load, which is why transcription got much faster in version 0.2.9.54.
Related terms
- Apple Neural Engine (ANE) ↗The dedicated ML accelerator inside every Apple Silicon chip, used to run ASR and diarization without burning the CPU or GPU.
- ANE (Apple Neural Engine) ↗Shorter name for the Apple Neural Engine - the per-chip ML accelerator that runs Mac Note Taker's transcription and diarization models.
- ASR (Automatic Speech Recognition) ↗The model layer that converts spoken audio into written text. The first stage of any meeting transcription pipeline.
- M-series (Apple Silicon) ↗Apple's family of ARM-based system-on-chip designs (M1 through M4) that power every modern Mac and ship with a Neural Engine.