surya758/wisprfree

WisprFree is a free, open-source dictation app for macOS that lives in your menu bar. It lets you speak to your Mac and get clean, polished text inserted directly into whatever app you are using. The developer built it out of frustration with dictation tools that transcribe every hesitation, false start, and misheard name, making raw speech-to-text barely usable for many people. The app works in two stages. First, a local speech recognition model running on your Mac listens to you and transcribes what you say. This step is fast and private because it happens entirely on your device. Second, an AI language model cleans up the transcript by fixing grammar, removing filler words, and applying a custom dictionary of names you have taught it. The cleaned text then lands wherever your cursor is. You can use it entirely offline with just the local transcription, or you can connect a cloud AI provider for the cleanup step. WisprFree supports several speech recognition engines, including NVIDIA's Parakeet models, OpenAI's Whisper Large, and Cohere Transcribe, covering multiple languages. For the AI cleanup, you bring your own provider: Google Vertex AI, Google Gemini, or any OpenAI-compatible endpoint such as OpenRouter, Groq, or a local Ollama instance. API keys are stored in the macOS Keychain, never in plain config files. The app offers three dictation modes you can switch between from the menu bar: Casual for light cleanup, Writing for aggressive prose polishing with your name dictionary applied, and Professional for business writing. Every mode prompt is editable so you can customize the behavior. A dictionary feature lets you add character and place names along with their common mishearings, so the cleanup model corrects them in context. You control recording with the Fn key: hold to talk, or use a hands-free mode. An overlay shows a live waveform while you speak. Output can be pasted, typed character by character, or copied to the clipboard. The app also tracks stats like words dictated and typing time saved. It requires Apple Silicon and macOS 14 or later.