opennookorg/betterwispr

BetterWispr – Offline Voice Dictation for macOS

What it is – BetterWispr is a free, open‑source macOS application that lets you dictate text into any window by holding a keyboard shortcut (default ⌥ Space). It runs entirely on your computer, using on‑device speech models (Apple’s built‑in recognizer, NVIDIA Parakeet, or OpenAI’s Whisper) so your voice never leaves the machine unless you deliberately connect an external API.

Key capabilities

• Hold‑to‑talk or toggle‑to‑talk – press the shortcut, speak, release and the transcribed text is pasted where your cursor is.
• On‑device models – Apple Speech (no download), Parakeet v2/v3 (25 European languages), Whisper large‑v3 turbo. All run offline after a one‑time download.
• Custom vocabulary – add spelling hints and phrase replacements for names, jargon, etc.
• Learning from corrections – fixing a mis‑heard word once (via History or immediately after paste) teaches the app to spell it correctly next time.
• Voice commands – say “comma”, “new line”, “scratch that”, “sorry, remove that”, or “at the rate KV” (→ “@KV”) to insert punctuation or symbols while dictating.
• Meeting mode – live transcription of both sides of a conversation, with optional on‑device titles, summaries and action‑item extraction using Apple Intelligence or a local Ollama model.
• History – raw and corrected transcripts are stored locally (can be disabled) for review or reuse.
• Bring‑your‑own model – connect any OpenAI‑compatible endpoint (e.g., Sarvam AI, Smallest AI, a self‑hosted server). Keys are kept in the macOS Keychain and audio is sent only to the chosen endpoint.

Privacy & data handling

• Settings, custom vocabularies and transcript history live in ~/Library/Application Support/BetterWispr/workspace.json (plain JSON, not encrypted).
• Model files are stored in a sibling Models folder.
• Audio recordings are written to temporary files and deleted immediately after each dictation.
• Built‑in models never contact the internet unless you explicitly download them; external model connections are only used when you enable them.

Getting started

1. Download the DMG from the latest release and drag BetterWispr to Applications.
2. Open the app, follow the welcome guide, and pick a model under Models.
3. Grant Microphone and Accessibility permissions (the latter lets the app paste text for you).
4. Click into any text field, hold ⌥ Space, speak, and release.
   • If pasting fails, the text is copied to the clipboard as a fallback.
   • Recordings auto‑stop after two minutes.

System requirements – macOS 14+ (Apple Silicon recommended).

Contributing – Build instructions, project layout, and contribution guidelines are in CONTRIBUTING.md.

License – Apache‑2.0 (third‑party model weights retain their own licenses).