Speech Output
Use speak only when the user asks to hear text aloud or is actively using a
voice conversation. Ordinary text replies do not need duplicate audio.
Method
- Keep spoken text concise and natural; remove Markdown syntax that would sound awkward.
- Call
speakwith the exact words to render. - Keep the normal text response available as the accessible transcript.
- If synthesis is disabled or unavailable, preserve the text and relay the exact setup or recovery step returned by the tool.
Boundaries
- Never enable voice output or microphone access on the user's behalf.
- Never treat a voice choice as consent to cloud processing; OSS speech output uses the configured local endpoint only.
- Never loop, replay, or start always-on listening from this tool.