README.md

---
language:
- vi
license: other
task_categories:
- audio-classification
- automatic-speech-recognition
pretty_name: ViMD Truncated 10s (16kHz)
size_categories:
- 10K<n<100K
---

# ViMD Truncated 10s — 16kHz

Preprocessed version of **ViMD** (Nguyen et al., EMNLP 2024) for Dialect Identification.

## Preprocessing applied
- **Resample:** 44.1kHz (original) → **16kHz, mono**
- **Truncate:** only the **FIRST 10 SECONDS** of each audio are kept
  (files shorter than 10s are kept intact). **1 original file = 1 sample.**
  This follows the truncation strategy of Lu et al. (2020), NOT chunking.
- **Splits:** original ViMD train/valid/test kept unchanged (speaker-exclusive).

## New column vs. original ViMD
- `new_province`: province name **after the administrative merger** (34 classes),
  mapped from the original `province` column (63 classes).

## Usage
```python
from datasets import load_dataset
ds = load_dataset("tannhoo06/ViMD_preprocessing")
```

## EDA without downloading audio
```python
import pandas as pd
meta = pd.read_parquet("hf://datasets/tannhoo06/ViMD_preprocessing/metadata.parquet")
```

## Citation
Please cite the original ViMD paper (Nguyen et al., EMNLP 2024).