Lindong-K/voice-input-plugin

README excerpt

Voice Input · 语音输入插件 A browser-side speech-to-text plugin for the DeepSeek Harness chat interface: click the mic, speak, and the transcript lands in the message box — no more typing long messages. Everything happens in your browser : microphone permission, recording, recognition, and writing into the input box. No server relay, no audio upload (unless you explicitly configure the optional Whisper endpoint). ✨ Features - A mic button injected right next to the send button (≤ 30px, rounded, tooltip "语音输入"). - Click to start listening: button turns red with a pulse animation, live interim results and a recognized-character counter appear in a preview strip under the composer. - Click again to stop: the final text is appended to the end of the input box (existing content kept, fully editable, input refocused). No auto-send unless you enable it. - Continuous dictation (Web Speech API continuous + interimResults ) — long monologues keep committing sentence by sentence, no 60s limit. - Cancel anytime (web-native, benchmarked against WeChat): while listening, a 「✕ 取消」 button appears in the preview strip and Esc discards the current session — including an automatic rollback of the text this…

Category