Text to Speech

Use the bundled scripts/speech.mjs from this directory. Node.js and the installed plugin's host-managed Gateway key are required. Do not install an SDK or ask for secrets in chat.

Workflow

1. Resolve the requested spoken text. For a sample of Junior's voice, use the bundled sample. Do not send unrelated private context or the full persona to the provider.
2. Use the default junior-v1 preset: a bright, warm, mischievous adult helper with dry humor. Use --delivery plain for serious or sensitive text. Do not present it as an imitation or recording of a real performer.
3. Write narration to a UTF-8 file. Keep each clip at 1–5000 characters. Generation spends Gateway credits. Do not generate extra variants without a request.
4. Generate to a fresh .wav path. Inspect an existing file before trying again. Do not automatically retry timeouts or failures: generation may already have spent credits.
5. Use sendFiles to deliver the returned path as audio/wav. Say it is synthetic narration. For a video, measure the new audio and pass it to the rendering workflow; speech generation alone does not update the video.

Commands and Assets

# Local, no API call: show the reusable voice preset.
node scripts/speech.mjs --profile