BetterWispr

BetterWispr is a free, open-source macOS dictation app built around local speech recognition. It lets a user hold Option Space in any app, speak naturally, and insert cleaned text into the current field when recording stops. The product is aimed at people who want faster writing without sending every spoken phrase through a transcription server. The official site describes the core workflow in plain terms: set up the app, grant permissions, install a local speech model, hold the shortcut, talk, and release to type. BetterWispr then removes filler words, repeated phrases, and false starts depending on the chosen cleanup level. It can also learn vocabulary such as product names and technical terms, so phrases like Postgres, TypeScript, or BetterWispr are spelled the way the user expects. Local-first privacy is the main differentiator. BetterWispr says models run on the Mac, no account is required, dictation history stays local, and offline use works after compatible assets are installed. It supports Apple Speech, Parakeet TDT v3 for 25 European languages, Parakeet Japanese, and Whisper Large v3 Turbo through WhisperKit. Optional providers can be configured for some workflows, but the baseline product is designed to avoid cloud transcription. The app also includes practical writing controls. Users can choose hold-to-talk or press-to-toggle behavior, change the keyboard shortcut, use voice commands such as question mark, new paragraph, and scratch that, and set app-specific writing styles. A note or email can be cleaned differently from a chat message. Meeting notes are built in as well: the site shows a short conversation turning into action items. BetterWispr is strongest for Mac users who write a lot in Slack, email, issue trackers, notes, docs, and code review comments. It is also useful for privacy-sensitive users who like dictation but do not want a hosted speech product. It is less appropriate for Windows users, managed enterprise fleets that require central policy controls, or people who need live word-by-word streaming because the product transcribes after recording stops. For builders, the appeal is simple: fast dictation, local models, no account, and an Apache 2.0 codebase that can be inspected or self-built. It also fits people who move between many writing surfaces because the same shortcut works in the active app instead of forcing every thought through a separate editor or browser tab. That makes the product feel like a system service rather than another inbox to manage. The tradeoff is that model downloads, permissions, and Mac hardware still matter, so setup is more involved than a hosted web recorder.