Speech generation

Turn text into WAV audio files in the agent folders, through OpenRouter (Gemini 3.8 Flash TTS or Flash Lite TTS). Uses the workspace OpenRouter key.

Slug: speech-generation Unlocks tools: generate_speech

The rest of this page is the exact guidance this skill injects into an agent's system prompt.

---

Speech generation

This tool group gives you generate_speech: it turns text into a WAV audio file written in one of your folders, and returns its path. (WAV, not mp3: the models answer raw audio.)

How to call it

• text: the exact words to speak, at most 5,000 characters. Everything in it is read aloud: no stage directions, no "(pause)", no "[cheerful]".
• style: the tone, in a few words ("warm and friendly", "calm and slow", "whispering"). It is not read aloud.
• path: where to write the file, ending in .wav (e.g. audio/intro.wav). Missing folders are created.
• model: google/gemini-3.8-flash-tts (default, the expressive tier) or google/gemini-3.8-flash-lite-tts (faster and cheaper, plain reading).
• voice: the voice name as Google names it; Kore is the default.

Discipline

1. Longer than 5,000 characters: split at paragraph boundaries into several files (part-1.wav, part-2.wav…) and say so in your answer.
2. Give the path back in your answer: the owner finds the file there.
3. A failure is said, not retried blindly. No OpenRouter key, an unknown voice, a provider error: report the reason the tool returned.