Hugging Face's logo

Datasets:
Overtech
/
SYRIA-GENSPEECH

Tasks:
Automatic Speech Recognition
Text-to-Speech
Modalities:
Audio
Text
Formats:
parquet
Languages:
Arabic
Size:
1K - 10K
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files
xet
 Community
SYRIA-GENSPEECH / README.md
Overtech's picture
Overtech
Update README.md
ff44912 verified
preview code
|
Raw History Blame Contribute Delete
9.57 kB
metadata 
language:
  - ar
pretty_name: 'SYRIA-GENSPEECH: Synthetic Syrian Arabic Speech Dataset'
homepage: https://huggingface.co/datasets/Overtech/SYRIA-GENSPEECH
license: cc-by-4.0
size_categories:
  - 1K<n<10K
task_categories:
  - automatic-speech-recognition
  - text-to-speech
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*.parquet
dataset_info:
  features:
    - name: id
      dtype: string
    - name: text
      dtype: string
    - name: audio
      dtype:
        audio:
          sampling_rate: null
  splits:
    - name: train
      num_examples: 3000

SYRIA-GENSPEECH

Synthetic Syrian Arabic Speech Dataset

An open speech resource by Overtech

SYRIA-GENSPEECH is a curated collection of 3,000 AI-generated Syrian Arabic speech recordings, each paired with its corresponding Arabic transcript.

Developed and published by Overtech, this dataset contributes to the availability of accessible Syrian Arabic speech resources for research, experimentation, and the development of speech and language technologies.

The collection is designed as a general-purpose resource rather than a benchmark restricted to a particular task. Its recordings are synthetically generated and should not be interpreted as recordings of natural human speech.

Dataset at a Glance

Attribute Details
Dataset SYRIA-GENSPEECH
Publisher Overtech
Language Arabic
Dialect Syrian Arabic
Speech origin AI-generated / Synthetic
Total recordings 3,000
Total audio duration 3 hours, 51 minutes, 18 seconds
Audio format WAV
Sampling rate 48,000 Hz
Channels Mono
Dataset size 1.69 GiB
Data format Parquet with embedded audio
Available split Train
License CC BY 4.0
Access Public

Overview

Syrian Arabic is a linguistically rich variety of spoken Arabic, with vocabulary, pronunciation patterns, and everyday expressions that are not always adequately represented in broadly available speech resources.

SYRIA-GENSPEECH was prepared to provide an additional source of paired Syrian Arabic text and synthetic speech, offering a structured collection that can be explored, processed, and incorporated into a variety of language-technology workflows.

The dataset emphasizes three principles:

Accessibility. Public availability in a standardized format, enabling researchers and developers to work with the collection without requiring a proprietary distribution process.

Practical usability. Direct alignment between each recording and its corresponding transcript, with self-contained audio storage and consistent metadata.

Transparency. Explicit disclosure of the synthetic origin of the recordings, their technical characteristics, and the limitations associated with generated speech.

Dataset Composition

The release contains 3,000 audio–text pairs organized into a single train split.

Each entry consists of a unique identifier, a Syrian Arabic transcript, and its associated audio recording.

Feature Schema

Field Type Description
id String Unique identifier assigned to the recording
text String Arabic transcript associated with the recording
audio Audio Corresponding WAV recording embedded within the dataset

Distribution Format

The dataset is distributed through six self-contained Parquet shards:

SYRIA-GENSPEECH/
├── README.md
├── LICENSE
└── data/
    ├── train-00000-of-00006.parquet
    ├── train-00001-of-00006.parquet
    ├── train-00002-of-00006.parquet
    ├── train-00003-of-00006.parquet
    ├── train-00004-of-00006.parquet
    └── train-00005-of-00006.parquet

The filenames above illustrate the six-shard layout.

Audio content is embedded within the Parquet records using a Hugging Face-compatible audio feature. Separate WAV file downloads or absolute local file paths are not required.

Audio Specifications

The following statistics were calculated from the embedded audio headers during release preparation.

Property Value
Encoding/container WAV
Sample rate 48 kHz
Channel configuration Mono
Shortest recording 1.760 seconds
Longest recording 8.320 seconds
Average recording 4.626 seconds
Total duration 03:51:18
Total recordings 3,000

These measurements describe the published audio collection and do not constitute an assessment of perceptual quality, pronunciation accuracy, or linguistic coverage.

Creation and Provenance

All speech recordings in SYRIA-GENSPEECH were generated synthetically using artificial intelligence and paired with their corresponding Syrian Arabic transcripts.

The release was prepared through a structured process of dataset organization, metadata preparation, packaging, and technical validation.

The published collection contains generated speech rather than a corpus of recorded human conversations.

No claim is made that the synthetic audio fully reproduces the pronunciation, prosody, emotional expressiveness, or natural variation of native Syrian Arabic speakers.

Intended Applications

SYRIA-GENSPEECH is provided as a flexible resource for exploration and development across speech and language technology.

Potential applications include:

  • Automatic speech recognition research and experimentation
  • Text-to-speech research and evaluation
  • Speech–text alignment and multimodal processing
  • Arabic dialect-focused language technology
  • Speech representation and audio feature analysis
  • Educational demonstrations and reproducible experiments
  • Synthetic speech analysis and benchmarking

These are potential applications, not claims of demonstrated performance or suitability for every task.

The dataset provides paired speech and text. It does not include additional ground-truth labels for emotion, speaker identity, demographic attributes, or audio-event classes.

Responsible Use and Limitations

Users should consider the following characteristics when evaluating or applying this resource.

Synthetic speech characteristics. Generated recordings may contain acoustic artifacts, pronunciation inconsistencies, or unnatural timing and intonation.

Dialect representation. The dataset is identified as Syrian Arabic, but comprehensive representation of all Syrian regional dialects and local linguistic varieties has not been established.

Speaker and demographic diversity. The collection should not be assumed to represent the full diversity of natural speakers, ages, genders, or speaking styles.

Evaluation suitability. Synthetic recordings should not automatically be treated as a substitute for naturally recorded speech when evaluating real-world speech recognition or generation systems.

Application-specific validation. Users are responsible for assessing data quality, bias, licensing requirements, and suitability for their intended applications.

The collection is distributed as a research and development resource without guarantees of fitness for a particular purpose.

License

SYRIA-GENSPEECH is published under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

Under this license, users may share, redistribute, and adapt the licensed material, including for commercial purposes, provided that the applicable license conditions are satisfied.

These conditions include appropriate attribution, reference to the license, and identification of modifications where required.

Full license terms are available in the repository's LICENSE file and at:

https://creativecommons.org/licenses/by/4.0/

© 2026 Overtech. Rights and permissions apply to the extent recognized and held under applicable law.

Attribution and Acknowledgment

Recognizing the Work Behind the Data

SYRIA-GENSPEECH represents an investment of time and effort in the creation, organization, preparation, validation, and publication of a publicly accessible Syrian Arabic speech resource.

By releasing this collection openly, Overtech aims to support researchers, developers, educators, and the wider speech technology community.

We value responsible reuse and respectfully encourage everyone building upon this resource to acknowledge its origin.

Attribution helps preserve the connection between derivative work and the original dataset, recognizes the effort involved in preparing the collection, and allows others to discover the source.

When sharing or adapting the licensed material, please follow the attribution conditions of CC BY 4.0.

Preferred attribution

Overtech — SYRIA-GENSPEECH: Synthetic Syrian Arabic Speech Dataset
https://huggingface.co/datasets/Overtech/SYRIA-GENSPEECH
Licensed under CC BY 4.0

Citation

If you use SYRIA-GENSPEECH in research, publications, software, demonstrations, or derivative datasets, please acknowledge the dataset using the following reference:

Overtech. (2026).
SYRIA-GENSPEECH: Synthetic Syrian Arabic Speech Dataset.
Hugging Face Datasets.
https://huggingface.co/datasets/Overtech/SYRIA-GENSPEECH

Where appropriate, include the dataset version or access date to identify the release used.

No DOI or associated academic publication is claimed for this release.

Published by Overtech · 2026

Supporting accessible resources for Syrian Arabic speech and language technology.