Murmur
Hold a key, speak, release. The words appear where your cursor is — transcribed on your Mac, never sent anywhere.
Download for Mac Requires macOS 14 or later on Apple Silicon. Signed and notarized. Free, MIT-licensed.How it works
Push-to-talk onlyHold your chosen key — a single modifier like right ⌥ works — and release to transcribe. No toggle mode to forget about.
On-device WhisperOpenAI's Whisper models run on the Neural Engine via WhisperKit. Small, Medium and Large v3 Turbo; English, Russian and more, with auto-detect.
Types where you areText goes straight into the focused field through the Accessibility API, with a clipboard-restoring paste as fallback. Works in Notes, Mail, browsers, Slack, editors, terminals.
Clean outputSentence capitalisation, optional filler-word removal, and a replacement dictionary for names and jargon the model mishears.
What it needs, and why
- Microphone — to record while the key is held. Audio stays in memory and is discarded after transcription.
- Accessibility — to notice the push-to-talk key while another app is in front, and to place text at your cursor. Murmur matches keyboard events against your hotkey only; it does not record keystrokes.
- Disk space — 0.5–1.5 GB per model, downloaded once from Hugging Face when you choose it.
Privacy
No cloud API, no account, no telemetry, no automatic update check. The only network use is the model download you start yourself (plus a one-time tokenizer fetch on first load). Full statement: PRIVACY.md.
Support
Problems or ideas: open an issue. Help → Save Diagnostics Report… in the menu writes a file to your Desktop that makes bug reports much easier to act on — it contains no dictated text.