FuzzySoul/dsh-chatvoice
Free voice closed loop for the Web UI: browser SpeechRecognition mic input with live interim results plus read-aloud speaker buttons and auto-read for assistant replies, zero configuration and no API key.
ChatVoice (dsh-chatvoice) is a free, zero-config, no-API-key voice plugin for DeepSeek Harness: dictate prompts and have assistant replies read aloud. Everything runs on the browser's native Web Speech API -- SpeechRecognition for input and speechSynthesis for read-aloud -- so there is no backend, no key, and no network request (the plugin spawns no subprocesses). A mic button in the composer toolbar appends each confirmed sentence to the input box in real time while you keep typing (speech only appends, never rewrites, so deleted text stays deleted), and a speaker button on every assistant reply reads it aloud; an optional auto-read reads new replies automatically. Settings cover recognition language (zh-CN/en-US), auto-read, voice, and rate.
Install
dsh plugin --profile web add dsh-chatvoicenpm package dsh-chatvoice 0.1.7 (registry re-verified 2026-09-30; repository field back-links to github.com/FuzzySoul/dsh-chatvoice, MIT). The README also documents a manual pnpm add dsh-chatvoice (the profile bundles reconcile automatically). Restart dsh web and open http://127.0.0.1:3080 -- the README warns you must access dsh web via 127.0.0.1 because speech recognition needs a secure context (over a LAN IP the microphone is blocked; read-aloud still works). Settings live under Settings -> ChatVoice and persist to ~/.dsh/chatvoice.json.
Compatibility
DeepSeek Harness Web UI, accessed through 127.0.0.1 (localhost secure context). The README recommends Edge (Xiaoxiao Online Natural Chinese voice; Azure-backed recognition that is more reliable in China than Chrome's Google-servers path); Firefox/Safari do not support SpeechRecognition (the mic button is disabled with a hint while read-aloud still works).
Details
- Repo: FuzzySoul/dsh-chatvoice
- Category: UI Enhancements
- Stars: 1
- Version: npm dsh-chatvoice 0.1.7 (registry re-verified 2026-09-30); MIT
- Last push: 2026-08-16
- First seen: 2026-08-16
Recent updates
The README documents the host/client split (host: config schema + GET/POST /dsh-chatvoice/config, settings persisted to ~/.dsh/chatvoice.json; client: MutationObserver-injected mic and speaker buttons), the friendly-error toasts for permission/unsupported/insecure-context/network cases, and a Phase 2 roadmap (push-to-talk, edge-tts voices, voice commands, voice memos, an agent-callable read_aloud tool).
FAQ
- How do I install dsh-chatvoice?
- Run dsh plugin --profile web add dsh-chatvoice (or pnpm add dsh-chatvoice), restart dsh web, and open http://127.0.0.1:3080. Per the README you must use 127.0.0.1 because the microphone needs a secure context.
- Does it need an API key?
- No. The README states it runs entirely on the browser's native Web Speech API -- no backend, no key, no registration, no network requests.
- Why is Edge recommended?
- Per the README, Edge's recognition goes through Azure (more reliable in China than Chrome's Google path) and Edge ships the most natural free Chinese voice, Xiaoxiao Online (Natural), which the plugin auto-picks.
Alternatives
forrestahha/dsh-voice-input · Jesse-njx/dsh-voice · Zhangbo-cn/dsh-voice-input-plugin