GooDAnDReaDY/dsh-voice

README 摘要

dsh-voice Voice input for the DeepSeek Harness Web GUI, in two modes, each with its own provider fallback chain. Dictation — press the mic, talk, and the text lands in the composer as you go: speech is cut into phrases on silence and each phrase is transcribed on its own. Press again to stop; sending stays in your hands. Voice message — press the wave button, record, press again. The transcript is inserted and sent to the agent after a short cancel window. Both modes fall back across providers, so one outage or rate limit does not lose your recording. API keys never reach the browser: audio is posted to the plugin's own route and the host talks to the providers. Install Restart the Web UI afterwards, then hard-refresh the browser. Providers Key Service Default model Credential browser the browser's own speech recognition — none, and nothing is uploaded to the host deepgram Deepgram nova-2 DEEPGRAM API KEY groq Groq whisper-large-v3-turbo GROQ API KEY hf HuggingFace Inference openai/whisper-large-v3 HF TOKEN local-whisper local whisper.cpp server model given at server start none, fully offline Keys are read through the DSH credentials service (Settings → Credentials, or $DSH HOME/.c…

分类