Jesse-njx/dsh-voice
Voice notes in, spoken answers out — dictate audio that becomes user messages (t
Voice notes in, spoken answers out — a hands-free terminal for DSH. Two tools and one durable event: transcribe({source}) does speech-to-text from an existing audio file or live mic recording; the transcript becomes a user message the agent responds to — never tool output — and the chat shows a compact audio card with play/pause, duration, backend badge, and transcript caption. The /voice command and readReplies toggle narrate the agent's replies aloud. Nothing runs until the model calls a tool.
Install
dsh plugin --profile web add github:Jesse-njx/dsh-voiceGitHub install (npm package @dsh-voice/bundle is not on the registry — 404 verified 2026-08-29): dsh plugin --profile web add github:Jesse-njx/dsh-voice. The bundle installs the dsh-voice entry (tools + /voice command + web audio cards). Config: stt.backend (whisper-local | openai | macos | fake), tts.backend (say | piper | edge-tts | fake), readReplies, audioDir.
Compatibility
DeepSeek Harness bundle. STT backends: whisper-local / openai / macos / fake; TTS backends: say / piper / edge-tts / fake.
Details
- Repo: Jesse-njx/dsh-voice
- Category: Messaging & Communication
- Stars: 2
- Version: GitHub source v0.1.0 (npm @dsh-voice/bundle 404 — not published)
- Last push: 2026-08-13
- First seen: 2026-08-13
Recent updates
transcribe tool (file or mic); audio cards with play/pause + backend badge; /voice command; spoken replies via readReplies; pluggable STT/TTS backends.
FAQ
- Which STT backends are supported?
- whisper-local, openai, macos, and fake — auto-selected (whisper-local → macos) when unset.
- Can it read replies aloud?
- Yes — the readReplies toggle narrates the agent's answers using the configured TTS backend (say / piper / edge-tts).
- Does it run constantly?
- No — nothing runs until the model calls a tool.
Alternatives
forrestahha/dsh-voice-input · Scorp1o117/dsh-soul-md · SaiSenBox/dsh-prompt-manager