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.

Log in or Sign Up to review the conditions and access this dataset content.

VieNeu-TTS-140h-nanocodec

Dataset tiếng Việt đã được encode bằng NVIDIA NeMo NanoCodec cho training Kani TTS.

Dataset Details

  • Source: VieNeu-TTS-140h (pnnbao-ump/VieNeu-TTS-140h)
  • Total samples: 74,858
  • Speakers: 193
  • Language: Vietnamese (vi)
  • Codec: NVIDIA NeMo NanoCodec (22kHz, 0.6kbps, 12.5fps)
  • Format: JSONL.gz với NanoCodec tokens (4 layers)

Dataset Structure

Mỗi mẫu bao gồm:

  • text: Text transcription tiếng Việt
  • nano_layer_1: Codec tokens layer 1
  • nano_layer_2: Codec tokens layer 2
  • nano_layer_3: Codec tokens layer 3
  • nano_layer_4: Codec tokens layer 4
  • encoded_len: Độ dài sequence tokens
  • speaker: Speaker ID
  • lang: "vi"
  • dataset_source: "vieneu-tts-140h"
  • phonemized_text: IPA phonemization (nếu có)
  • gender: Giới tính speaker (nếu có)

Usage

from datasets import load_dataset

dataset = load_dataset("pnnbao-ump/VieNeu-TTS-140h-nanocodec")

# Xem một mẫu
sample = dataset['train'][0]
print(sample['text'])
print(f"Speaker: {sample['speaker']}")
print(f"Tokens: {sample['encoded_len']}")

License

CC BY-NC 4.0 - Non-commercial use only

Citation

@dataset{vieneu_tts_140h_nanocodec,
  title = {VieNeu-TTS-140h NanoCodec Dataset},
  author = {Pham Nguyen Ngoc Bao},
  year = {2025},
  url = {https://huggingface.co/datasets/pnnbao-ump/VieNeu-TTS-140h-nanocodec}
}
Downloads last month
90

Models trained or fine-tuned on pnnbao-ump/VieNeu-TTS-140h-nanocodec