phoonnx-chatterbox-lahgtna

phoonnx-chatterbox-lahgtna

Mirror of Lahgtna, a Chatterbox fine-tune specialized for spoken Arabic dialects, exported to ONNX for phoonnx, OpenVoiceOS's offline TTS library. It is fine-tuned from Chatterbox Multilingual and shares its architecture: an autoregressive Llama-backbone codec-LM with zero-shot cloning from an untranscribed reference clip and an exaggeration control. See docs/training/engines/chatterbox.md and docs/cloning.md.

Ten dialect codes are exposed as separate voice IDs, one per Arabic-speaking region: Egypt (eg), Saudi Arabia (sa), Morocco (mo), Iraq (iq), Lebanon (lb), Sudan (sd), Libya (ly), Syria (sy), Tunisia (tn), and Palestine (ps). All map to the same underlying language_model_q4.onnx graph in this repository; the dialect selection happens in text normalization/tokenization, not a different model per dialect.

Related repositories: phoonnx-chatterbox (base English), phoonnx-chatterbox-turbo (faster GPT-2-based variant), phoonnx-chatterbox-multilingual (the base this was fine-tuned from).

Usage

Cloning support requires pip install phoonnx[cloning].

import wave
from phoonnx.model_manager import TTSModelManager

manager = TTSModelManager()
manager.load()
manager.merge_default_voices()

voice = manager.voices["chatterbox/lahgtna/eg"].load()
with wave.open("out.wav", "wb") as wav_file:
    voice.synthesize_wav(
        "أهلاً، إزيك؟",
        wav_file,
        speaker_reference="reference.wav",  # optional: clone this speaker
    )

License

Lahgtna is released by its authors (oddadmix) under the MIT license, the same license as the Chatterbox base and multilingual models it descends from. phoonnx itself is Apache-2.0.