Local speech-to-text and text-to-speech skills for Claude Code.
| Skill | Description |
|---|---|
| ez-stt | Local speech-to-text using Parakeet (best English accuracy) or Whisper (fastest, multilingual) |
| ez-tts | Local text-to-speech using Kyutai Pocket TTS with 8 preset voices and voice cloning |
Both skills run fully offline after initial model download. TTS supports english, french, german, portuguese, italian, and spanish. A Hugging Face token (HF_TOKEN) is needed for model download / voice cloning (accept terms on https://huggingface.co/kyutai/pocket-tts).
npx skills add araa47/ez-voiceThis installs both ez-stt and ez-tts skills.
# Transcribe audio (Parakeet, best English accuracy)
scripts/stt.py audio.ogg
# Use Whisper backend (faster, multilingual)
scripts/stt.py audio.ogg -b whisper# Generate speech with default voice
uv run scripts/tts.py "Hello, world"
# Choose a voice / language
uv run scripts/tts.py "Hello, world" --voice marius
uv run scripts/tts.py "Hola mundo" --language spanish
# Output as OGG
uv run scripts/tts.py "Hello, world" --oggSee each skill's SKILL.md for full usage details.
- Install dependencies:
uv sync --all-extras - Make your changes
- Ensure pre-commit hooks pass:
prek run --all-files - Ensure tests pass:
uv run -m pytest - Submit a PR