MayBog/bla

ChatterboxTTS Hebrew Fine-tuned Model

This is a Hebrew fine-tuned version of ResembleAI/chatterbox for Hebrew text-to-speech synthesis.

Model Details

• Base Model: ResembleAI/chatterbox
• Language: Hebrew
• Training Data: Hebrew phonemes dataset
• Fine-tuning: T3 component only (voice encoder and S3Gen frozen)
• Final Training Loss: 7.8869

Usage

from transformers import AutoModel
import torch

# Load the fine-tuned model
model = AutoModel.from_pretrained("MayBog/orpheus-hebrew-lora-v3")

# Use with ChatterboxTTS pipeline
# (Add specific usage instructions based on your implementation)

Training Details

• Training Steps: 4373
• Batch Size: 4 (per device)
• Learning Rate: 5e-05
• Gradient Accumulation: 2 steps
• Warmup Steps: 100
• Evaluation Strategy: Every 2000 steps

Model Architecture

This model uses the ChatterboxTTS architecture with:

• Voice Encoder: Frozen (pre-trained weights)
• S3Gen: Frozen (pre-trained weights)
• T3: Fine-tuned on Hebrew data

Dataset

Trained on a Hebrew phonemes dataset with audio-phoneme pairs for Hebrew text-to-speech.

Limitations

• Optimized specifically for Hebrew language
• May not perform well on other languages
• Fine-tuned only on T3 component

Citation

@misc{chatterbox-hebrew-finetuned,
    title={ChatterboxTTS Hebrew Fine-tuned Model},
    author={[Your Name]},
    year={2025},
    url={https://huggingface.co/MayBog/orpheus-hebrew-lora-v3}
}