Arabic Orpheus TTS Fine-tuning

1. Multi-speaker Arabic adaptation
2. Single-speaker specialization

Highlights

• Formal Arabic / MSA-oriented output
• 2-stage LoRA fine-tuning
• Arabic text normalization
• SNAC-based audio tokenization
• Demo audio samples included

Important note on data

• raw audio
• metadata
• transcriptions
• processed dataset artifacts derived from the source corpus

Training setup

• Base model: unsloth/orpheus-3b-0.1-ft
• Stage 1: Arabic multi-speaker training
• Stage 2: single-speaker fine-tuning on a female voice
• Hardware used: L40S
• LoRA fine-tuning with Unsloth

🔊 Audio Demos

Demo 1

Demo 2

Demo 3

Notes

Citation

@misc{toyin2025arvoicemultispeakerdatasetarabic,
      title={ArVoice: A Multi-Speaker Dataset for Arabic Speech Synthesis}, 
      author={Hawau Olamide Toyin and Rufael Marew and Humaid Alblooshi and Samar M. Magdy and Hanan Aldarmaki},
      year={2025},
      eprint={2505.20506},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2505.20506}, 
}