araa47 ez-tts - Local Text-to-Speech
Local text-to-speech using Kyutai's Pocket TTS. Fully offline after initial model download. ~10x realtime on CPU, no GPU needed.
Supports english, french, german, portuguese, italian, spanish (pocket-tts 2.x).
Usage
# Basic English (outputs to /tmp/tts_output.wav)
uv run scripts/tts.py "Hello, this is a test"
# Specify preset voice
uv run scripts/tts.py "Hello world" --voice marius
# Other languages (default voice picked per language)
uv run scripts/tts.py "Hola mundo" --language spanish
uv run scripts/tts.py "Bonjour le monde" --language french
uv run scripts/tts.py "Ciao mondo" --language italian --voice giovanni
# Custom output file
uv run scripts/tts.py "Hello world" --out ~/my_audio.wav
# Output as OGG (for voice messages, requires ffmpeg)
uv run scripts/tts.py "Hello world" --ogg
# Voice cloning from a local sample (requires HF_TOKEN + HF access to kyutai/pocket-tts)
uv run scripts/tts.py "Hello world" --voice /path/to/sample.ogg
Languages
| Language | Notes | Default voice |
|---|---|---|
english (default) |
Same as english_2026-04 |
alba |
english_2026-01 / english_2026-04 |
English model variants | alba |
spanish / spanish_24l |
_24l = higher quality, slower |
lola |
german / german_24l |
juergen |
|
italian / italian_24l |
giovanni |
|
portuguese / portuguese_24l |
rafael |
|
french_24l |
French is 24-layer only (french alias works) |
estelle |
Preset Voices
English: alba, marius, javert, jean, fantine, cosette, eponine, azelma, anna, vera, charles, paul, george, mary, jane, michael, eve, bill_boerst, peter_yearsley, stuart_bell, caro_davy
Other languages: giovanni (it), lola (es), juergen (de), rafael (pt), estelle (fr)
Custom cloned voices can be placed in the skill's voices/ directory and used by name (e.g., --voice myvoice if voices/myvoice.ogg exists).
Voice Cloning
Clone any voice from a clear speech sample:
uv run scripts/tts.py "Hello world" --voice ~/voices/my_voice.ogg
Requirements:
- Accept the terms on https://huggingface.co/kyutai/pocket-tts
HF_TOKENenvironment variable set- Audio sample should be clear speech (WAV/OGG/MP3/FLAC)
Preset catalog voices work without cloning weights; file-path cloning needs the full model.
Options
--voice/-v— Voice name or path to audio for cloning (default depends on--language)--language/-l— Model language (default:english)--out/-o— Output file path (default:/tmp/tts_output.wav)--ogg— Convert output to OGG (requires ffmpeg)--quiet/-q— Suppress progress messages
Environment Variables
| Variable | Purpose |
|---|---|
HF_TOKEN |
Hugging Face token for model download / voice cloning |