DSH Live Voice

Local-first speech-to-text (STT), text-to-speech (TTS), and meeting audio transcription plugin for DeepSeek Harness (DSH). Local speech processing does not make DSH or its AI provider fully offline.

Installation

Tested with DeepSeek Harness v0.2.1-rc.2. The package manifest's dshTestedVersion field records the tested version.

Install the prebuilt package from npm; no source build is required. The Git repository contains source only.

npx dsh plugin add --profile web dsh-live-voice

Restart DSH, then open DSH Settings → Live Voice to configure recognition and speech engines.

Requirements and configuration

• Microphone capture requires browser support, permission, and a secure context (HTTPS or localhost).
• Recognition and synthesis engines have separate language, platform, and model requirements. See the engine guide before selecting a setup.
• Meeting transcription uses Qwen HTTP or Whisper HTTP recognition. Shared-audio availability depends on the browser, system, and sharing surface; obtain appropriate participant consent.
• Capturing meeting audio and sharing assistant playback require separate setup. This is not a native Google Meet or Teams integration.
• The optional dsh-live-voice-debugger package is not required by the main plugin.

Documentation

• Project overview — canonical presentation, screenshots, and use cases.
• Configuration and conversation flow
• Speech recognition engine selection
• Architecture
• Release notes
• Issue tracker

License

Apache-2.0