---
license: mit
task_categories:
- automatic-speech-recognition
- text-to-speech
language:
- ur
tags:
- urdu
- phonemes
- g2p
- speech
- ipa
size_categories:
- 10K<n<100K
---

# Urdu ONYX WAV Annotated Dataset

This dataset is an enhanced version of `humair025/Urdu-ONYX-WAV-real` with added phoneme annotations using the `urdu-g2p` library.

## Dataset Description

- **Total Samples**: 66,882
- **Language**: Urdu (اردو)
- **Voice**: Onyx
- **Annotations**: IPA phonemes with and without stress markers

## Features

| Column | Description |
|--------|-------------|
| `id` | Sample ID |
| `transcript` | Original Urdu transcript |
| `voice` | Voice type (onyx) |
| `text` | Text content |
| `timestamp` | Recording timestamp |
| `audio` | Audio file (WAV format) |
| `phonemes` | Space-separated IPA phonemes with stress markers |
| `phonemes_no_stress` | Space-separated IPA phonemes without stress |
| `phonemes_list` | List of individual phonemes |
| `phoneme_count` | Number of phonemes in the text |

## Usage

```python
from datasets import load_dataset

# Load dataset
dataset = load_dataset("humair025/Urdu-ONYX-WAV-Annoted")

# Access a sample
sample = dataset['train'][0]
print(f"Text: {sample['text']}")
print(f"Phonemes: {sample['phonemes']}")
print(f"Count: {sample['phoneme_count']}")

# Listen to audio
from IPython.display import Audio
Audio(sample['audio']['array'], rate=sample['audio']['sampling_rate'])
```

## Processing Details

- Phoneme conversion performed using [urdu-g2p](https://pypi.org/project/urdu-g2p/)
- Processing date: 2026-01-20
- Batch size: 2000 rows per batch
- Memory-efficient processing: 3 arrow files at a time
- Format: Arrow (preserves audio for playback in dataset viewer)

## Source

Original dataset: [humair025/Urdu-ONYX-WAV-real](https://huggingface.co/datasets/humair025/Urdu-ONYX-WAV-real)

## Citation

If you use this dataset, please cite the original source and the urdu-g2p library.

```bibtex
@misc{urdu-onyx-wav-annotated,
  title={Urdu ONYX WAV Annotated Dataset},
  author={Humair},
  year={2025},
  publisher={HuggingFace},
  url={https://huggingface.co/datasets/humair025/Urdu-ONYX-WAV-Annoted}
}
```

## License

MIT License