README.md

---
language:
- vi
- en
license: cc-by-4.0
task_categories:
- automatic-speech-recognition
pretty_name: AI2Text Bilingual ASR Dataset (Vietnamese + English)
size_categories:
- 100K<n<1M
---

# AI2Text – Bilingual ASR Dataset

A large-scale bilingual (Vietnamese + English) speech dataset used to train
the `Cong123779/AI2Text-Bilingual-ASR` model.

## Dataset Summary

| Split | Samples | Notes |
|-------|---------|-------|
| train | ~194,167 | 77% Vietnamese, 23% English |
| val   | ~30,123 | held-out validation |

## Data Fields

Each `manifest.csv` has the following columns:

| Column | Description |
|--------|-------------|
| `id` | Unique sample identifier |
| `transcript` | Ground-truth text (prefixed with `<\|vi\|>` or `<\|en\|>`) |
| `audio_path` | Relative path to the `.wav` file |
| `duration` | Duration in seconds |
| `words_json` | JSON array of word-level timestamps |

## Audio Format

- Sample rate: **16,000 Hz**
- Channels: **Mono**
- Format: **WAV**

## Language Distribution

- **Vietnamese** (`<|vi|>` token): ~77%
- **English** (`<|en|>` token): ~23%

## Usage

```python
import pandas as pd
from datasets import load_dataset

# Load only the manifest CSVs (fast, no audio)
ds = load_dataset("Cong123779/AI2Text-Bilingual-ASR-Dataset",
                  data_files={"train": "train/manifest.csv",
                               "val":   "val/manifest.csv"})
print(ds)
```

## License
Creative Commons Attribution 4.0 (CC-BY 4.0)