Anilosan15/Dia-1.6B-tr

Dia-1.6B (Turkish Full Fine-tuned)

This model is a fully fine-tuned Turkish version of Dia-1.6B, adapted specifically for Turkish language speech synthesis using custom multi-speaker data.

Unlike LoRA or adapter-based approaches, this model was fully fine-tuned, allowing deeper adaptation of pronunciation, prosody, and language modeling for Turkish.

---

🔊 Training Data

The training data was collected from publicly available YouTube channels and carefully processed (segmentation, cleaning, transcript alignment).

🎤 Speakers

Speaker Tag	Source Channel	Gender	Duration
sıla	https://www.youtube.com/@BirDinle	Female	~33 hours
taha	https://www.youtube.com/@seslifikiristasyonu	Male	~28 hours

• Total duration: ~61 hours
• Language: Turkish (tr-TR)
• Domain: Narration / voice-over style speech
• Segment length: ~3s – 20s

Speaker tags are explicitly used during training to enable multi-speaker inference.

---

🧠 Training Details

• Base model: Dia-1.6B
• Training type: Full fine-tuning (all parameters updated)
• Objective:
  • Strong adaptation to Turkish language
  • Improved pronunciation and phoneme consistency
  • Natural prosody for long-form Turkish narration
  • Preservation of speaker identity via speaker tags
• Data source: Publicly available YouTube content

---

📜 License & Usage Notice (Academic / Non-Commercial)

This model is released strictly for academic research and experimental purposes.

• The model is not intended for commercial use.
• Commercial deployment, resale, or use in revenue-generating products or services is explicitly prohibited without prior permission.
• The training data was derived from publicly accessible online sources and was used for research and educational objectives only.
• Users are responsible for ensuring compliance with applicable laws, platform terms, and ethical guidelines when using this model.

By using this model, you agree that it is provided “as-is”, without warranties of any kind, and solely for non-commercial academic use.