dsh-agent-voice
Give your DeepSeek Harness agent a voice. The agent speaks aloud when it finishes a task and when it needs your attention — never its reasoning, progress, or interim reports.
- Local, offline TTS (Piper voices + XTTS v2), no cloud API required.
- A curated set of voices (Czech + English) from rhasspy/piper-voices.
- Nice settings panel in the main interface, with live preview.
- Automatic announcements when subagents / workflows settle, and when the agent asks for a decision.
🇨🇿 Český návod →
Install
dsh plugin add dsh-agent-voice
Restart DSH, then open the 🔊 Agent Voice tab in the main interface.
How it works
| Action | When | Example |
|---|
dsh_agent_voice action='speak' text='…' | A task completed | "Úkol dokončen." |
dsh_agent_voice action='attention' | The agent needs you | "Potřebuji tě u počítače." |
The agent is instructed to never read its reasoning or intermediate steps aloud, and to keep spoken messages to one sentence.
Voices
| id | Voice | Language |
|---|
jirka | Jirka (medium) | cs |
kasandra | Kasandra (medium) | cs |
lessac | Lessac (medium) | en |
ryan | Ryan (high) | en |
xtts | XTTS v2 (multilingual) | cs, en |
Voices are downloaded from HuggingFace on first use (Piper .onnx models) into ~/.local/share/piper/.
License
MIT. Piper voice models are MIT-licensed (rhasspy/piper-voices). XTTS v2 is subject to Coqui's non-commercial CPML license; set COQUI_TOS_AGREED=1 to agree (the plugin does this automatically).