Training Artifacts

This folder contains reproducibility notes and the training hyperparameters used to produce the published model.

khmer_tts_training.yaml is the project-level training config.
finetune_mms_khm.json is the generated VITS fine-tuning config with dataset paths replaced by placeholders.

The raw training audio, internal automation scripts, and local intermediate training states are intentionally excluded from the Hugging Face release folder. Exact reproduction requires a Khmer audio/text dataset with the same format.

To fine-tune further, replace the placeholder dataset paths with your own Hugging Face AudioFolder dataset or a compatible local dataset.