README.md

metadata

language:
  - ps
license: mit
pretty_name: Pashto Synthetic Speech Dataset Parquet (5k)
size_categories:
  - 10K<n<100K
tags:
  - audio
  - speech
  - pashto
  - tts
  - synthetic
  - parquet
task_categories:
  - automatic-speech-recognition
metrics:
  - wer
paperswithcode:
  eval_results: []
  preferred_citation: |-
    @dataset{pashto_synthetic_speech_parquet_5k,
      title = {Pashto Synthetic Speech Dataset Parquet (5k)},
      year = {2025}
    }

Pashto Synthetic Speech Dataset Parquet (5k)

This dataset contains 10000 synthetic speech recordings in the Pashto language, with 5000 male voice recordings and 5000 female voice recordings, stored in Parquet format.

Dataset Information

• Dataset Size: 5000 sentences
• Total Recordings: 10000 audio files (5000 male + 5000 female)
• Audio Format: WAV, 24kHz, 16-bit PCM, embedded directly in Parquet files
• Dataset Format: Parquet with 500MB shards
• Sampling Rate: 24kHz (24000 Hz)
• Voices:
  • Male: ps-AF-GulNawazNeural
  • Female: ps-AF-LatifaNeural
• Total Audio Duration:
  • Male: 4310.24 seconds
  • Female: 4396.27 seconds
• Average Clip Duration:
  • Male: 0.86 seconds
  • Female: 0.88 seconds

Dataset Structure

pashto_speech_5k/
  ├── data/                     # Parquet data files with embedded audio
  │   ├── train-00000-of-....parquet
  │   ├── train-00001-of-....parquet
  │   ├── ...
  │   ├── _schema.json          # Parquet schema definition
  │   └── _index.json           # Parquet metadata
  ├── dataset_info.json         # Dataset metadata
  └── README.md                 # This file

Usage with HuggingFace Datasets

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("your-username/pashto_speech_5k")

# Access an example
example = dataset['train'][0]
print(example['sentence'])            # The transcript
print(example['audio'])           # Audio data
print(example['speaker'])         # Speaker gender
print(example['sampling_rate'])   # Audio sampling rate

Usage Applications

This dataset can be used for:

• Training ASR (Automatic Speech Recognition) systems for Pashto
• Voice activity detection
• Speaker verification
• Speech synthesis evaluation

License

This dataset is licensed under the MIT License.

Citation

If you use this dataset in your research, please cite:

@dataset{pashto_synthetic_speech_parquet_5k,
  title = {Pashto Synthetic Speech Dataset Parquet (5k)},
  year = {2025},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/datasets/ihanif/pashto_speech_5k}}
}

Acknowledgments

This dataset was created using Microsoft Edge TTS voices.