fcitx5-vinput

fcitx5-vinput adds voice typing to Fcitx5 on Linux. It's for people who want to dictate text and use spoken instructions to edit it, with a choice between local speech recognition and cloud services.

• Shortcut-based dictation: Hold a key while speaking and release it to finish, or tap to start and stop recording.
• Local or cloud recognition: Transcribe speech offline with sherpa-onnx models, or use Doubao, Aliyun Bailian, ElevenLabs, and OpenAI-compatible speech services.
• Text correction and rewriting: Use LLM processing to correct transcription errors, format text, translate it, or rewrite it according to an instruction.
• Spoken editing commands: Select existing text and speak an instruction to modify it, so voice input can handle edits as well as dictation.
• Graphical and command-line controls: Use the GUI to manage models and settings, or the CLI for control over models, providers, rewriting scenes, and the background service.

The local sherpa-onnx recognition path works offline. Cloud recognition sends audio to the selected provider and requires an internet connection; LLM rewriting is a separate processing capability from offline transcription.

The project is open source under GPL-3.0 and requires Fcitx5. Linux packages are available for Arch, Fedora, Ubuntu/Debian, openSUSE, Nix, and Flatpak.