Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Pocket-TTS generates speech from text using the Kyutai model locally on CPU, with no GPU required. It supports voice cloning from short audio samples and streaming output with under 200ms latency. English only, approximately 6x faster than real-time on modern hardware.
Runs entirely on CPU without cloud API dependencies, making it faster to set up and cheaper to operate than hosted TTS services for local or offline workflows.
When to use it
- Narrating documentation without a cloud TTS subscription
- Cloning a custom voice for consistent audio branding
- Generating speech offline on a laptop during travel
- Prototyping voice interfaces without API rate limits
- Batch-converting written content to audio files
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: generate speech from a script using my own voice sample
- 1Export voice embedding from audio sample with pocket-tts export-voice voice.wav voice.safetensors
- 2Load the TTSModel and retrieve voice state from the safetensors file
- 3Call model.generate_audio with the voice state and input text
- 4Save the returned audio array as a 24kHz WAV file using scipy
A 24kHz mono WAV file with speech in the cloned voice
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.