liangyongxiang@hotmail.com Yongxiang Liang vowstar@gmail.com Huang Rui fcitx5-vinput is a voice input addon for the Fcitx5 input method framework. It performs on-device speech recognition with offline sherpa-onnx models and can optionally route the recognized text through any OpenAI-compatible LLM endpoint for error correction, formatting and translation. Cloud ASR providers can also be added at runtime through a downloadable registry. The package ships the fcitx5 addon, a background daemon, a command line client and a Qt6 graphical configuration tool. The prebuilt sherpa-onnx and ONNX Runtime shared libraries are bundled privately. fcitx5-vinput 是 Fcitx5 输入法框架的语音输入插件。它使用离线的 sherpa-onnx 模型在本地完成语音识别，并可选择把识别结果发送到任意 兼容 OpenAI 接口的大语言模型端点，用于纠错、格式化和翻译。云端 ASR 服务也可以通过可下载的注册表在运行时添加。本软件包包含 fcitx5 插件、 后台守护进程、命令行客户端以及基于 Qt6 的图形配置工具，并私有打包了 预编译的 sherpa-onnx 与 ONNX Runtime 共享库。 xifan2333/fcitx5-vinput