CoolFace
Datasetpublic

shekar-ai/Neyshekar

Neyshekar Neyshekar is an open, community-driven Persian speech dataset collected via a web-based crowdsourcing platform at https://ney.shekar.io. It is designed to support research and development in text-to-speech (TTS), automatic speech recognition (ASR), speech representation learning, and other downstream Persian speech applications. The recordings are provided by volunteer contributors, all of whom are native Persian speakers. Each release represents a… See the full description on the dataset page: https://huggingface.co/datasets/shekar-ai/Neyshekar.

sourceHugging Facecc0-1.0updated 8d agoView on Hugging Face
2likes1.2kdownloads
Dataset Card

<p align="center"> <img src="https://huggingface.co/datasets/shekar-ai/neyshekar/resolve/main/header.png" alt="Neyshekar" width="80%"> </p>

<p align="center"> <a href="https://arxiv.org/abs/2609.14542"><img src="https://img.shields.io/badge/arXiv-2609.14542-b31b1b.svg" alt="arXiv"></a> <a href="https://t.me/shekarai" target="blank"><img src="https://telegram-badge.vercel.app/api/telegram-badge?channelId=@shekar_ai&color=00A693" alt="Telegram" /></a> </p>

Neyshekar

Neyshekar is an open, community-driven Persian speech dataset collected via a web-based crowdsourcing platform at [https://ney.shekar.io](https://ney.shekar.io). It is designed to support research and development in text-to-speech (TTS), automatic speech recognition (ASR), speech representation learning, and other downstream Persian speech applications.

The recordings are provided by volunteer contributors, all of whom are native Persian speakers. Each release represents a stable snapshot of the dataset, enabling reproducible research and consistent benchmarking.

Audio is mono, 16-bit PCM, sampled at 16 kHz.

Latest Release - V6.0 (2026-09-01)

  • —Total samples: 62,279
  • —Informal samples: 15,222 (24.44%) (identified using Shekar rule-based InformalClassifier)
  • —Total duration (hours): 99.02
  • —Average clip duration (seconds): 5.72
  • —Total tokens: 701,621
  • —Vocab size: 29,535
  • —Speakers: 190

Splits

V6.0 is the first release to ship predefined splits. Splits are assigned per speaker, so a single recorder never straddles two sets and the evaluation splits stay speaker-disjoint from train. The speaker_id column makes this checkable: no identifier appears in more than one split.

SplitSamplesShareHoursSpeakers
train58,24493.5%91.99134
validation1,8863.0%3.1426
test2,1493.5%3.8830
python
from datasets import load_dataset

dataset = load_dataset("shekar-ai/neyshekar")
train, validation, test = dataset["train"], dataset["validation"], dataset["test"]

Fields

  • —id: Unique sample identifier
  • —audio: Speech recording
  • —text: Persian transcription
  • —duration: Clip duration in seconds
  • —speaker_id: Anonymous, stable identifier for the recorder

Links

Terms of Use

Any attempt to identify or uncover the identity of speakers is strictly prohibited.

License

This dataset is released under the CC0 1.0 Universal license. It may be used, modified, and redistributed for any purpose without restriction.

Citation

If you use Neyshekar in your research, please cite:

bibtex
@article{amirivojdan2026neyshekar,
  title={Neyshekar: An Open Persian Read-Speech Corpus for Automatic Speech Recognition},
  author={Amirivojdan, Ahmad and Nadiri, Farzad and Alizadeh, Abolfazl and Yaraghi, Shaghayegh},
  journal={arXiv preprint arXiv:2609.14542},
  year={2026},
  doi={10.48550/arXiv.2609.14542},
  url={https://arxiv.org/abs/2609.14542}
}