KotDath/dsh-voice

README 摘要

dsh-voice Voice input for the DeepSeek Harness Web GUI — as an official plugin bundle . A microphone button (SVG icon) sits next to the send button in the composer. Clicking it brings up a ChatGPT-style recording "pill" above the composer: - ✕ — cancel the recording; - ⏹ — stop, transcribe and insert the text into the input (without sending); - ↑ — stop, transcribe and send to the chat immediately. The waveform is a row of vertical bars: fresh signal arrives on the right, history fades left into a dim dotted baseline (like ChatGPT's voice mode). Installation Requirements - A running DeepSeek Harness with the Web GUI (Node.js quick start: npx @deepseek-ai/dsh web , default http://127.0.0.1:3080 ). - On the host machine: bash , curl (engine/model downloads), ffmpeg (audio conversion). The transcribe.cpp engine is downloaded automatically per OS/architecture from this repository's Releases — or adopted from PATH / a sibling .engine/ dir if a transcribe-cli binary already exists. - A browser with mic access over HTTPS or localhost. Install into a profile (all sessions) For a local test build instead: Then restart the web process ( dsh web ) — the new plugin row is picked up at boot. Af…

分类