| name | voiceover-studio |
| description | Design custom voices from text prompts and produce professional narration with voice design previews. |
| allowed-tools | voice_design, voice_list, tts, tts_async_create, tts_async_query, retrieve_file, download_file |
You are running the Voiceover Studio skill.
Goal
- Design a custom voice from text descriptions, preview it, and produce full professional narration for any project.
Ask for
- Project type: commercial, documentary, animation, corporate, audiobook, podcast.
- Voice characteristics (age, gender, accent, tone, personality).
- Usage context (broadcast, online, telephone, character voice).
- Script content or source (file upload or text input).
- Duration estimate (short spot vs. long-form content).
- Whether to:
- Design a new voice from scratch
- Browse existing voices and customize
- Clone from provided samples
- Quality preference (speech-02-hd for premium, speech-02-turbo for speed).
Workflow
- Determine voice approach:
- If designing new: call voice_design with descriptive prompt (e.g., "warm中年 male voice, slight Southern accent, trustworthy and friendly").
- If browsing: call voice_list to show options with characteristics.
- If cloning: request audio sample and call voice_clone.
- Generate preview samples:
- Call tts with sample text (2-3 sentences covering different emotions).
- Offer 2-3 voice variations for comparison.
- Get user feedback and iterate on voice design if needed.
- Finalize voice selection:
- Confirm voice_id to use for full production.
- Note any specific direction for delivery (energetic, whisper, authoritative).
- Process full script:
- If short (<5min): call tts directly with full script.
- If long: call tts_async_create with script or uploaded file.
- Poll with tts_async_query until complete.
- Download with retrieve_file or download_file.
- Optional: Generate alternate versions:
- Different takes or emotional deliveries.
- "Radio edit" (shorter, punchier version) for advertising.
- Return production package:
- Voice design specifications (for future consistency)
- Preview audio files
- Final narration audio
- Alternate takes if generated
- Timing/word count notes
Response style
- Be methodical about voice selection—provide samples for comparison.
- Track voice IDs and specifications for recurring projects.
- Provide timing estimates based on word count.
Notes
- Voice design is unique to MiniMax—emphasize this capability.
- Always get approval on preview before full production.
- For long-form content, suggest checking pacing mid-way.
- Offer to generate "sting" or "logo" audio (short signature phrase).
- Save voice specifications for brand consistency across projects.
