BetterWispr

BetterWispr is a free, open-source voice dictation app for macOS and a privacy-first alternative to cloud tools like Wispr Flow. Hold Option+Space in any app, speak naturally, and release: the transcript is cleaned up and typed where your cursor was. Speech recognition runs on your Mac with Apple's on-device speech, NVIDIA Parakeet (English, Japanese and 25 European languages) or Whisper large-v3 turbo, so audio is not uploaded and models keep working offline once downloaded. There is no account and no cloud fallback.

The cleanup layer is what sets it apart from plain speech-to-text. Light cleanup removes ums, repeats and false starts; medium cleanup also edits for clarity using Apple Intelligence or a local Ollama model. You can set a different tone per app (casual in Messages, formal in email), say voice commands such as "comma", "new paragraph" or "scratch that", and add names and jargon as spelling hints. Fix a misheard word once and BetterWispr remembers it. A built-in meeting notetaker transcribes you and the other side separately and writes summaries and action items on-device, and a local history keeps raw and corrected text (you can switch it off).

It needs macOS 14 or later (Apple Silicon recommended) and offers optional bring-your-own-model connections to Sarvam AI, Smallest AI or any OpenAI-compatible endpoint. Source code and releases are on GitHub.