README.md

---
license: mit
language:
- ur
tags:
- tts
- urdu
- speech-synthesis
- audio
task_categories:
- text-to-speech
size_categories:
- 10K<n<100K
---

# XCollab/urdu-tts

## Dataset Description

Combined Urdu Text-to-Speech dataset with **17,575 high-quality audio-text pairs**.

## Dataset Structure

```python
{
    'index': int,           # Sample index
    'text': str,           # Urdu text transcription  
    'audio': {             # Audio data
        'path': str,        # Path to audio file
        'sampling_rate': 16000
    },
    'source_dataset': str  # Original dataset name
}
```

## Usage

```python
from datasets import load_dataset
import soundfile as sf

# Load dataset
dataset = load_dataset("XCollab/urdu-tts")

# Get a sample
sample = dataset['train'][0]
print("Text:", sample['text'])
print("Audio path:", sample['audio']['path'])

# Save audio file
sf.write("sample.wav", sample['audio']['array'], sample['audio']['sampling_rate'])

# Play in notebook
import IPython.display as ipd
ipd.Audio(sample['audio']['array'], rate=sample['audio']['sampling_rate'])
```

## Audio Specifications
- **Sampling Rate**: 16,000 Hz
- **Format**: Float32 arrays
- **Channels**: Mono
- **Quality**: Normalized and validated

## Statistics
- **Total Samples**: 17,575
- **Valid Samples**: 17,575
- **Processing Errors**: 0

## Source Datasets
This dataset combines samples from multiple Urdu TTS datasets for improved coverage and quality.

## License
Please refer to individual dataset licenses for usage terms.