Chatterboxegy v0

--- language: - ar tags: - text-to-speech - arabic - egyptian-arabic - chatterbox - finetuned base_model: resemble-ai/chatterbox --- # Chatterbox — Egyptian Arabic Finetune This is a finetuned version of [Chatterbox TTS](https://github.com/resemble-ai/chatterbox) on Egyptian Arabic dialect speech. ## Training Details - **Base model**: resemble-ai/chatterbox (ChatterboxMultilingualTTS) - **Dataset**: MAdel121/arabic-egy-cleaned (2000 samples) - **Epochs**: 12 - **Language**: Egyptian Arabic (`ar`) - **Finetuned layers**: Prosody/duration predictors + late acoustic decoder layers ## Usage ```python from chatterbox.mtl_tts import ChatterboxMultilingualTTS import torch model = ChatterboxMultilingualTTS.from_pretrained(device="cuda") # Load finetuned weights ckpt = torch.load("t3_state_dict.pt", map_location="cuda") model.t3.load_state_dict(ckpt, strict=False) wav = model.generate("أهلاً وسهلاً", language_id="ar") ```