Datasets:

Tnaot
/

SPS-Bopha-Voice-Dataset-v1

Tasks:

Modalities:

Formats:

Languages:

Size:

Tags:

Libraries:

License:

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

This dataset is formatted for fine-tuning VibeVoice.

training_data.jsonl: The main manifest file containing transcriptions and paths.
chunks_staging/: Directory containing the audio clips.

Clone this repository:

git clone https://huggingface.co/datasets/Tnaot/SPS-Bopha-Voice-Dataset-v1
cd SPS-Bopha-Voice-Dataset-v1

Run the training script pointing to training_data.jsonl. The paths in the JSONL file are relative (e.g., chunks_staging/file.wav), so they will resolve correctly as long as you run the trainer from the root of this directory.
```
python -m src.finetune_vibevoice_lora        --train_jsonl training_data.jsonl        ...
```