README.md

---
language:
- dv
license: apache-2.0
task_categories:
- automatic-speech-recognition
- text-to-speech
tags:
- audio
- dhivehi
- speech
- president
- political
size_categories:
- 1K<n<10K
---

# Dhivehi Presidential Speech Dataset (Parquet Format)

This is a parquet-formatted version of the [dash8x/dv-presidential-speech](https://huggingface.co/datasets/dash8x/dv-presidential-speech) dataset, converted from the old loading script format to modern parquet format for better compatibility.

## Dataset Summary

Dhivehi Presidential Speech is a Dhivehi speech dataset containing around 2.5 hours (1 GB) of speech collected from Maldives President's Office consisting of 7 speeches given by President Yaameen Abdhul Gayyoom.

## Dataset Structure

### Data Splits

| Split | Utterances | Duration |
|-------|------------|----------|
| Train | 1,612 | ~2h 15m |
| Validation | 200 | ~17m |
| Test | 200 | ~13m |

### Data Fields

- `audio`: Audio object with:
  - `array`: Audio waveform as numpy array
  - `sampling_rate`: 16,000 Hz
  - `path`: Original file path
- `sentence`: Dhivehi transcription text

## Usage

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("Serialtechlab/dv-presidential-speech")

# Access a sample
sample = dataset['train'][0]
print(f"Text: {sample['sentence']}")
print(f"Audio shape: {sample['audio']['array'].shape}")
print(f"Sampling rate: {sample['audio']['sampling_rate']} Hz")
```

## Supported Tasks

- **Automatic Speech Recognition (ASR)**: Train models to transcribe Dhivehi speech
- **Text-to-Speech (TTS)**: Train models to synthesize Dhivehi speech

## Language

Dhivehi (Maldivian) - `dv`

## License

Apache 2.0

## Citation

Original dataset: [dash8x/dv-presidential-speech](https://huggingface.co/datasets/dash8x/dv-presidential-speech)

Data extracted and processed by Sofwath from the [DhivehiDatasets](https://github.com/Sofwath/DhivehiDatasets) collection.