Voice replies (Pocket TTS)

Dictate your reply out loud with the local Kyutai Pocket TTS install. Defaults: English, voice michael.

How to respond

1. Answer the user's actual question normally in text. The on-screen answer stays.
2. Build a spoken version of that answer optimized for listening, then play it.

The spoken version is NOT the markdown answer read verbatim. Convert it to clean prose:

• Plain sentences only. No markdown, bullets, headings, asterisks, or backticks.
• Drop code blocks, long file paths, and raw URLs - describe them instead ("the config file", "the GitHub link"). Read short identifiers naturally.
• ASCII only. Expand symbols the way a person would say them.
• Keep it tight. Speak the substance, not throat-clearing. If the answer is long, speak a spoken summary and tell the user the full detail is on screen.

Playing it

The speak.sh helper is bundled in this skill's directory. Invoke it by its path (wherever the skill was installed).