Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Qwen3-TTS VoiceDesign converts text to speech using natural language voice descriptions combined with seed-based timbre control. The same description and seed pair always produces the same voice, enabling reproducible custom voices. Ships with an OpenAI-compatible HTTP server, one-click setup, and batch seed exploration scripts.
Seed-based timbre fixation means the same description and seed pair always reproduces the exact same voice, unlike most TTS APIs where voice consistency depends on proprietary speaker IDs.
When to use it
- Generating voiceovers for videos with a consistent custom voice
- Adding local TTS to an OpenClaw agent via OpenAI-compatible endpoint
- Exploring voice timbres by batch-testing dozens of random seeds
- Producing Chinese-language speech with specific accents or broadcast styles
- Self-hosting a TTS API to avoid third-party cloud service dependencies
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: generate a deep male broadcaster voice saying 'Welcome to the news'
- 1Set TTS_INSTRUCT to a description like '30岁男性播音员，声音低沉磁性，语速稳重从容'
- 2Run batch_seeds.sh to generate samples across 30-40 seeds
- 3Listen to samples and shortlist 2-3 candidates
- 4Set the chosen seed as TTS_SEED in .env
- 5Call say.sh with the input text to produce the final audio file
An MP3 file of 'Welcome to the news' spoken in the selected broadcaster voice
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.