Skip to content

NoteScribe is a focused macOS voice transcription app for automatic speech recognition. It runs fully offline using Parakeet TDT Frontier Core ML audio models for minimal impact to system resources while maximizing transcription speed.

Notifications You must be signed in to change notification settings

swaylenhayes/NoteScribe

Repository files navigation

NoteScribe

NoteScribe is a focused macOS voice transcription app for automatic speech recognition. It runs fully offline using Frontier Core ML audio models from Fluid Inference (Nvidia Parakeet TDT v2 and v3) and is designed for fast dictation into any text field or for transcribing audio/video files.

There are two single-model builds, plus one combined build:

  • Parakeet V2 (English-only)
  • Parakeet V3 (Multilingual - 25 European languages)
  • Parakeet V2+V3 Combined (both models bundled; larger download)

If you only need English, choose V2 since it has a slightly lower WER for English than V3. If you need multilingual transcription, choose V3. If you want in-app model switching between V2 and V3 without separate installs, choose the combined build.

Latest Release

  • Version: v1.3.3 (February 18, 2026)
  • Highlights: refreshed start/stop/paste/escape sounds, a fix for unintended Escape cue playback during normal stop, and an added combined V2+V3 release artifact.
  • Release notes: /Users/swaylen/dev/NoteScribe/_docs/RELEASE_NOTES_v1.3.3.md
  • Downloads: GitHub Releases

Model performance (M-series optimized)

Both models are Core ML models optimized for Apple silicon. In practice, Parakeet V2/V3 Core ML runs among the fastest offline ASR options on Mac today. Compared to Whisper GGML (base / large-v3 / turbo), these models show minimal extra RAM overhead, similar CPU thread usage, and a clear shift of work to the Neural Engine, leaving the rest of the system more responsive.

benchmark_graph_paraCoreML_vs_whisperGGML_harvard100_swaylenhayes_12-10-2025

Compared to GGML models, the CoreML models use:

  • ~50% less RAM,
  • ~50% less energy draw,
  • ~30-40% less CPU load (at peak)

What it does

  • Hotkey dictation that pastes into the frontmost app
  • Scratchpad tab for dumping and editing text
  • File transcription (audio/video)
  • History and replacements

Install

  1. Download the DMG for v2, v3, or v2v3 (combined) from GitHub Releases.
  2. Open the DMG and drag NoteScribe.app into Applications.
  3. Launch the app and grant microphone (and accessibility for paste) when prompted.

Uninstall

An uninstall helper script is included in this repo: uninstall_notescribe.sh.

Requirements

  • macOS 14+
  • Apple silicon (M-series). Intel Macs are not supported.

Developer Notes

  • Build/sign/notarize commands: /Users/swaylen/dev/NoteScribe/_docs/DEV_BUILD_COMMANDS.md
  • Current implementation + release packaging status: /Users/swaylen/dev/NoteScribe/_docs/CURRENT_STATE.md

About

NoteScribe is a focused macOS voice transcription app for automatic speech recognition. It runs fully offline using Parakeet TDT Frontier Core ML audio models for minimal impact to system resources while maximizing transcription speed.

Resources

Stars

Watchers

Forks

Packages

No packages published

Contributors 2

  •  
  •