Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
Minor documentation update
What This Skill Does
Runs text-to-speech and speech-to-text entirely on Apple Silicon using FluidAudio's CoreML models via the Apple Neural Engine. TTS uses the Kokoro model with 54 voices at ~0.6-0.8s latency; STT uses Parakeet TDT v3 with ~0.2-0.3s latency across 25 languages. No network connection or API key required.
Achieves sub-second voice synthesis and transcription on-device, replacing cloud TTS/STT services without API costs, rate limits, or sending audio data off-device.
When to use it
- Reading long articles aloud while multitasking
- Transcribing recorded meetings without uploading audio to the cloud
- Adding offline voice output to a local AI assistant
- Converting written notes to audio files for listening on the go
- Building voice interfaces for local automation scripts
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Read this paragraph aloud
- 1Receives text input from the user
- 2POSTs text to the local TTS endpoint at http://127.0.0.1:18790/synthesize
- 3StellaVoice daemon processes the request using the Kokoro model on the Apple Neural Engine
- 4Returns a WAV file in approximately 0.6-0.8 seconds
- 5Plays the audio output to the user
The paragraph is spoken aloud in the selected Kokoro voice with natural pacing
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.