---
language: vi
license: other
tags:
- tts
- vietnamese
- audio
- speech-dataset
pretty_name: Vietnamese TTS Multi-Project Dataset
task_categories:
- text-to-speech
---

# Vietnamese TTS Multi-Project Dataset

A comprehensive Vietnamese Text-to-Speech (TTS) dataset comprising 118GB of high-quality audio recordings and aligned transcripts from multiple narrative literature projects.

## Dataset Structure

The data is organized into three main projects, matching the original folder structure for portability:

### 1. **The Gioi Hoan My (Thế Giới Hoàn Mỹ)**
- **Archives**: `the_gioi_hoan_my.tar.zst` (Split into parts due to size).
- **Metadata**: Includes aligned transcripts, scores, and training/validation splits.

### 2. **An Sat (Án Sát)**
- **Archives**: `an_sat.tar.zst`
- **Metadata**: Aligned CSV files and training text.

### 3. **Ngao The Cuu Trong Thien (Ngạo Thế Cửu Trọng Thiên)**
- **Archives**: `Ngao_The_Cuu_Trong_Thien_Phong_Lang_Thien_Ha_mono22050.tar.zst` (Split into parts).
- **Metadata**: Processed mono (22.05kHz) scripts and scoring metadata.

## File Reassembly Instructions

Due to Hugging Face's 50GB individual file limit, some large archives have been split into multiple parts. Please reassemble them using the following command after downloading:

```bash
# For The Gioi Hoan My
cat the_gioi_hoan_my.tar.zst.part_* > the_gioi_hoan_my.tar.zst

# For Ngao The Cuu Trong Thien
cat Ngao_The_Cuu_Trong_Thien_Phong_Lang_Thien_Ha_mono22050.tar.zst.part_* > Ngao_The_Cuu_Trong_Thien_Phong_Lang_Thien_Ha_mono22050.tar.zst
```

After reassembling, you can decompress the `.tar.zst` files using:
```bash
tar --use-compress-program=zstd -xvf <filename>.tar.zst
```

## Metadata Format

All `metadata_aligned.csv` files follow the standard TTS format (LJSpeech style):
`wav_path | transcript`

Example:
`wavs/001_1003_1012.wav|sát khí khôn cùng ùn ùn kéo đến bắn ra mãnh liệt!`

---
**Maintenance**: Managed by [Cong123779](https://huggingface.co/Cong123779).