Native macOS Floating Pill Preview (Click to test tone)
Wisper Emotion for macOS
Offline dictation for macOS. Hold fn, speak, and your words land in whatever app you are in.
Install with one command
Copy it, open Terminal (⌘ + Space → "Terminal"), paste, press Return. It only saves the installer to Downloads. Full steps
Nothing leaves your Mac·~500 MB RAM while dictating·Every model deletable
Supported Speech Models
Base downloads itself on first launch. Switch or delete any model from Settings — they download on demand.
Good balance of speed and accuracy
Quick notes and short messages
Technical words and names
Strong accents or noisy rooms
Other languages, slowest to respond
One-time download each. Only the selected model is ever loaded into memory.
Emotion detection is optional
Dictation never touches the tone model. Add it only if you want it, remove it whenever.
~142 MB on disk, ~500 MB RAM. Full transcription accuracy — the live overlay still tags Calm, Focused, Thoughtful or Energetic from pace and energy alone, at no cost.
Download it from Settings and finished utterances get a label — Neutral, Happy, Angry, Sad — with the confidence spread. One button removes it again.
The honest limit: tone from audio is right about 62–75% of the time — it hears delivery, not meaning. Anything under 50% confidence is marked unsure rather than shown as fact. All of it runs on your Mac.
Installation
The app is not signed with an Apple Developer ID, so macOS blocks browser-downloaded copies. curl never sets that quarantine flag — nothing to clear afterward.
Already have a blocked copy from somewhere else? Clear the quarantine flag:
Core Capabilities
Real-time voice dictation, acoustic emotion metrics, and instant system cursor injection.
"Drafting the architectural overview for our new Electron desktop engine. Wisper automatically captures vocal nuance and injects text into VS Code..."
FAQ
Installing, Gatekeeper, models, and what stays on your Mac.
Wisper Emotion runs as a native Electron desktop application on your Mac. It connects to an ultra-fast local C++ Whisper engine and acoustic neural classifier. Pressing fn or ⌘ ⌥ Space transcribes your voice and types text into your active editor or app.