Wisper

Wisper is a voice dictation app for people who want to speak into their desktop apps while keeping speech processing on their own computer. It remembers the window where dictation began and inserts the transcript there, even if you switch windows while talking. It can't return to a specific browser tab within that window.

It's free and open source under the MIT license, with no account or telemetry. It runs on Linux with X11 or Wayland, Windows, and macOS, including Apple Silicon and Intel Macs. Local transcription works fully offline once you've downloaded a model.

Its ONNX speech engines include Whisper, Parakeet, IndicConformer, Moonshine, SenseVoice and Qwen3. You can choose a smaller model or one with broader language coverage. Automatic language detection and language priorities support multilingual dictation.

Optional AI cleanup polishes the transcript without answering what you said. It can use Ollama, your own endpoint, or cloud services including OpenAI, Anthropic, Groq and OpenRouter. Cloud speech recognition is also optional through OpenAI, Groq, Sarvam or an OpenAI-compatible endpoint. Cloud transcription sends audio to the selected provider; cloud cleanup sends the transcript for processing.

Custom vocabulary corrects recurring terms and abbreviations. Push-to-talk and toggle recording work through a global shortcut, with a small overlay for recording feedback. Searchable history lets you replay audio, edit text or transcribe it again with another model. You control retention and whether Wisper keeps recordings.