Reference
Glossary,
in plain English.
The vocabulary of local-first Mac meeting transcription in 2026. ASR, diarization, voice embeddings, ScreenCaptureKit, Core ML, Neural Engine, Ollama. Every term has its own page, cross-linked, no marketing.
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.
Related: Apple Neural Engine (ANE), M-series (Apple Silicon), Edge AI
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.
Related: ANE (Apple Neural Engine), M-series (Apple Silicon), On-device, Edge AI
Apple's speech-to-text APIs
The transcription APIs built into macOS - useful, free, and not built for multi-speaker meeting audio.
Related: ASR (Automatic Speech Recognition), Core ML, Whisper, Parakeet TDT
ASR (Automatic Speech Recognition)
The model layer that converts spoken audio into written text. The first stage of any meeting transcription pipeline.
Related: Parakeet TDT, Whisper, VAD (Voice Activity Detection), ANE (Apple Neural Engine)
avconferenced
The macOS system process behind FaceTime and other audio/video calls. It often shows up as the thing holding your microphone.
Related: Core Audio process tap, ScreenCaptureKit
BlackHole
An open-source virtual audio driver that routes Mac audio between apps. The old answer to capturing system audio, largely replaced by ScreenCaptureKit.
Related: ScreenCaptureKit, Loopback, Core Audio process tap
Core Audio process tap
The macOS 14.2+ API that lets an app subscribe to another running app's audio output stream, with user permission.
Related: ScreenCaptureKit, BlackHole
Core ML
Apple's on-device model runtime. It decides whether a model runs on the Neural Engine, the GPU, or the CPU.
Related: Apple Neural Engine (ANE), ANE (Apple Neural Engine), ASR (Automatic Speech Recognition), M-series (Apple Silicon)
Edge AI
The pattern of running ML inference on the end-user's device rather than in a centralized data center.
Related: On-device, Apple Neural Engine (ANE), ANE (Apple Neural Engine), M-series (Apple Silicon)
End-to-end encryption (E2EE)
A property where only the communicating endpoints can read the content - the service operator cannot. Common in messaging, rarely true for meeting notetakers.
GDPR
EU regulation governing personal-data processing. Recording a meeting requires a lawful basis and disclosure; local-only storage simplifies residency.
Related: HIPAA, End-to-end encryption (E2EE), On-device
HIPAA
US law governing protected health information. Cloud notetakers handling clinical audio require a Business Associate Agreement; local-only flows do not.
Related: GDPR, End-to-end encryption (E2EE), On-device
LLM (Large Language Model)
A transformer-based text-generation model. Used in Mac Note Taker for summaries, action-item extraction, and speaker rename suggestions.
Related: Ollama, OpenAI-compatible endpoint, On-device
Loopback
Rogue Amoeba's paid virtual audio routing app. Powerful, but for meeting transcription it is overkill compared to ScreenCaptureKit.
Related: BlackHole, ScreenCaptureKit
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.
Related: Apple Neural Engine (ANE), ANE (Apple Neural Engine), Edge AI
Ollama
A local LLM runtime for macOS, Linux, and Windows. The default backend for Mac Note Taker's on-device AI summaries.
Related: LLM (Large Language Model), OpenAI-compatible endpoint, On-device
On-device
Processing that happens entirely on the user's hardware - no cloud, no network round-trip, no third-party data processor.
Related: End-to-end encryption (E2EE), Edge AI, HIPAA, GDPR
OpenAI-compatible endpoint
Any HTTP API that mimics OpenAI's /v1/chat/completions shape, allowing one client to talk to many backends with no code change.
Related: LLM (Large Language Model), Ollama
Parakeet TDT
NVIDIA's open-source ASR model family. Parakeet TDT v3 is Mac Note Taker's default English transcription model on the Neural Engine.
Related: ASR (Automatic Speech Recognition), Whisper, ANE (Apple Neural Engine)
RTFx (real-time factor)
How many times faster than real time a model transcribes. RTFx 30 means a 30-minute meeting takes one minute.
Related: ASR (Automatic Speech Recognition), Parakeet TDT, Whisper, Core ML
ScreenCaptureKit
Apple's modern macOS API for capturing screen video and system audio without a kernel extension or a virtual audio cable.
Related: Core Audio process tap, BlackHole, Loopback
Sortformer
NVIDIA's streaming diarization model: it labels who is speaking as the audio arrives, instead of after the recording ends.
Related: Speaker diarization, Voice embeddings, ASR (Automatic Speech Recognition)
Sparkle (auto-update)
The open-source framework that handles in-app software updates for unsigned-by-the-App-Store macOS apps like Mac Note Taker.
Related: On-device
Speaker diarization
The process of segmenting an audio stream into per-speaker turns - the 'who spoke when' problem.
Related: Voice embeddings, ANE (Apple Neural Engine), ASR (Automatic Speech Recognition)
VAD (Voice Activity Detection)
A lightweight model that flags which slices of audio contain human speech, used to gate the heavier ASR and diarization stages.
Related: ASR (Automatic Speech Recognition), Speaker diarization
Voice embeddings
A fixed-length numeric fingerprint of a voice, used to match the same speaker across multiple recordings.
Related: Speaker diarization, On-device
WER (word error rate)
The standard accuracy score for speech recognition: the share of words the model gets wrong. Lower is better.
Related: ASR (Automatic Speech Recognition), Whisper, Parakeet TDT
Whisper
OpenAI's open-source ASR model. Whisper Large v3 is Mac Note Taker's high-accuracy multilingual option.
Related: ASR (Automatic Speech Recognition), Parakeet TDT, ANE (Apple Neural Engine)
Read the longer pieces
The glossary is the reference. The field notes go deeper on diarization, on-device LLMs, and the 2026 privacy framework.