Creative Claw — Chatterbox

Read shared execution guidance once per task before using tools. It covers existing authorization, model discovery, optional cost checks, imports, and recovery.

Use this specialist inside creativeclaw-generate-voiceover for one-off reference-audio voice matching. Chatterbox is not a reusable Character voice: choose creativeclaw-clone-voice when the user wants to save and reuse a voice through ElevenLabs.

Consent boundary

Use a reference only when the user owns the voice or the speaker authorized this use. Do not infer consent from a public URL, interview, film, song, celebrity clip, or uploaded third-party recording. If authorization is unclear, stop before generation and ask for it.

Workflow

1. Preserve the user's script and confirm the supplied recording is authorized for voice matching.
2. Prefer a clean 5–30 second sample with one speaker, natural speech, little room echo, and no music or effects. A sample close to the target emotion and pace usually transfers better.
3. Import the sample through platform upload guidance so audio_url is a permanent, publicly retrievable Creative Claw URL.
4. Call get_model_params({ model: "speech/chatterbox" }); the current route is the original reference-audio TTS endpoint, not the separately documented multilingual or Turbo endpoints.
5. Call generate_speech with model: "speech/chatterbox", text, audio_url, and only runtime-supported settings.
6. Audition speaker similarity, pronunciation, pace, emotion, artifacts, hallucinated sounds, and clipped boundaries. Regenerate the smallest weak section.