officialuser/chatterbox-nepali

🇳🇵 Chatterbox-TTS (Nepali Edition)

This repository contains the fine-tuned training checkpoints for the Nepali language, based on the Chatterbox-Multilingual-500M model by Resemble AI.

🛠️ Implementation & Code

The full training logic, bug fixes for the Devanagari script, and the custom inference UI are hosted on GitHub. You must use the custom implementation to run these models correctly.

👉 GitHub: officialuser/chatterbox-nepali

📦 Model Contents

This repository provides:

• t3_nepali_epoch_20.pt: A robust intermediate checkpoint. You can use this for quick testing or as a base to continue training on your own Nepali datasets.
• t3_mtl_nepali_final.safetensors (User Generated): Once you train for the remaining epochs using the provided GitHub scripts, you will generate this high-performance production file.

🚀 Quick Usage

To test this model, follow the instructions in the GitHub repository:

1. Clone the repo: git clone https://github.com/officialuser/chatterbox-nepali.git
2. Install dependencies.
3. Download t3_nepali_epoch_20.pt from this page and place it in the root folder.
4. Run the test script:

python3 test_nepali.py \
  --checkpoint "t3_nepali_epoch_20.pt" \
  --ref_audio "your_nepali_sample.wav" \
  --text "नमस्ते, म नेपाली एआई हुँ।" \
  --output "output.wav"

⚖️ License & Acknowledgments

This model is released under the MIT License.

• Base model architecture (Chatterbox) by Resemble AI.
• Nepali fine-tuning and Devanagari script optimization by officialuser.
• Built for the open-source Nepali speech synthesis community.