SnowfallC/dsh-mobile-voice-call
App-free mobile voice calls with existing DeepSeek Harness sessions.
DSH Mobile Voice Call lets you scan a QR code with a phone or tablet and call the DeepSeek mascot directly from the browser -- no app to download. The plugin connects through DSH's supported plugin composition without patching the core: DSH, an authenticated bridge, whisper.cpp, and Hojo-TTS-Light run on the computer, a Cloudflare Quick Tunnel supplies a temporary HTTPS address, the phone browser captures audio and plays results, recordings return to the computer for local transcription, and the text is sent through DSH's native session RPC for a spoken reply. The phone page lists existing idle top-level sessions (hiding archived ones), requests microphone access on session select, and supports muting, interrupting, hanging up, and cancelling an active turn; it follows DSH's English/Chinese and light/dark preferences and keeps the one-time pairing token in the URL fragment then an HttpOnly SameSite=Strict cookie.
Install
pnpm dsh plugin --profile web add "github:SnowfallC/dsh-mobile-voice-call" && pnpm dsh webRun the two commands above from the DSH source repository. Requirements: a working DSH Web profile, Node/pnpm, and uv on PATH for Hojo-TTS-Light (a separately managed system Python is not required); the default automatic setup targets Windows x64. A local-checkout form is also documented. Uninstall with pnpm dsh plugin --profile web remove dsh-mobile-voice-call then restart DSH. The npm name dsh-mobile-voice-call was NOT found on the registry on 2026-09-30, so use the GitHub-source form.
Compatibility
A working DSH Web profile; the default automatic setup targets Windows x64. Speech recognition (whisper.cpp) downloads an ~8 MB Windows x64 build plus the multilingual small model (~466 MiB) with fixed sizes and SHA-256 digests verified; speech synthesis (Hojo-TTS-Light 80M) uses uv to create an isolated Python 3.12 environment, installs pinned CPU inference packages, and downloads ~348 MB of model files plus the pinned upstream script. The phone page must be HTTPS (a Cloudflare Quick Tunnel supplies it) and the browser must support microphone capture and Web Audio; recent Android Chrome and iOS Safari are recommended. Calls are turn-based, not full-duplex.
Details
- Repo: SnowfallC/dsh-mobile-voice-call
- Category: uncategorized
- Stars: 1
- Version: GitHub source install (npm name dsh-mobile-voice-call NOT published -- registry empty 2026-09-30); MIT
- Last push: 2026-08-17
- First seen: 2026-08-17
Recent updates
The README documents the requirements and the two install forms, the first-time speech-recognition and TTS setup (downloads, SHA-256 verification, cache under $DSH_HOME/cache/dsh-mobile-voice-call/), the browser requirements and limits (HTTPS, system browser not an embedded QR-scanner browser, local whisper.cpp recognition, turn-based calls, no end-to-end encryption, running sessions disabled in the selector), the local speech configuration options (whisperExecutablePath/whisperModelPath/whisperThreads; hojoReferenceAudioPath/hojoReferenceText/hojoThreads/hojoModelDirectory/hojoPythonPath), the sentence-level streaming behaviour, and an experimental warning about Cloudflare tunnels.
FAQ
- How do I install DSH Mobile Voice Call?
- Per the README, from the DSH source repository run pnpm dsh plugin --profile web add "github:SnowfallC/dsh-mobile-voice-call" then pnpm dsh web. A working DSH Web profile and uv on PATH are required; the default automatic setup targets Windows x64.
- Do I need to install an app on my phone?
- No -- the README says there is no app to download: scan a QR code with a phone or tablet and call the mascot directly from the browser (use a system browser, not an embedded QR-scanner browser).
- Is the audio private?
- Per the README, speech recognition runs locally through whisper.cpp, so raw recordings are not submitted to DeepSeek, but they still travel phone-to-computer through the Cloudflare tunnel, so it is not an end-to-end encrypted path; Quick Tunnels provide no stable hostname and should be revoked after use.
Alternatives
PandaPolo/dsh-voice-call · FuzzySoul/dsh-chatvoice · Richard-Yang0130/dsh-web-speech-input