chenxie95/gigaspeech3

GigaSpeech 3

Scaling Multi-Dialect Arabic Speech Recognition with Weakly Labeled Data

GigaSpeech 3 is a large-scale Arabic speech recognition corpus containing approximately 10 million weakly transcribed speech segments across five dialects: Algerian, Egyptian, Iraqi, Saudi, and Syrian Arabic. It supports research on dialect-specific and multi-dialect ASR, weak supervision, and low-resource speech recognition.

• Dataset: GigaSpeech 3
• Evaluation benchmark: GigaSpeechBench
• Related datasets: GigaSpeech, GigaSpeech 2

Dataset Description

Code	Dialect
DZA	Algerian Arabic
EGY	Egyptian Arabic
IRQ	Iraqi Arabic
SAU	Saudi Arabic
SYR	Syrian Arabic

Following the accompanying manuscript, dialect-specific labeling models are built using VietASR-style ASR-biased self-supervised learning and approximately 51–63 hours of manually transcribed speech per dialect. These models generate weak transcripts for dialect-matched web audio. Data selection and four rounds of noisy-student refinement for SAU and SYR produce the study's final training inventories.

The manuscript reports 23.86% macro-average WER for the final dialect-specific Zipformers on the five Arabic dialects of GigaSpeechBench, and evaluates transfer to MASC, Casablanca, and SADA. MSA speech used in the manuscript's multi-dialect experiments is separate from this five-dialect release.

Dataset Structure

Audio is packaged in tar shards, with a separate Parquet metadata file for each dialect:

GigaSpeech3/
├── README.md
├── DZA/
│   ├── audio/001.tar, 002.tar, ...
│   └── metadata/DZA.parquet
├── EGY/
├── IRQ/
├── SAU/
└── SYR/

All dialect directories follow the same layout. Each Parquet row contains:

Field	Description
id	Segment identifier
duration	Duration in seconds
text	Automatically generated transcript
dialect	Dialect code
shard	Audio tar filename, such as 001.tar
member	Audio path inside the tar archive

Each dialect is exposed as a train configuration. Human-transcribed evaluation data are hosted separately in GigaSpeechBench.

Example Usage

pip install -U datasets huggingface_hub

Load metadata without downloading the audio archives:

from datasets import load_dataset

REPO_ID = "Antares97/GigaSpeech3"
metadata = load_dataset(REPO_ID, "SAU", split="train", streaming=True)
row = next(iter(metadata))
print(row)

Download the archive referenced by a metadata row:

from huggingface_hub import hf_hub_download

archive_path = hf_hub_download(
    repo_id=REPO_ID,
    repo_type="dataset",
    filename=f"{row['dialect']}/audio/{row['shard']}",
)

The corresponding audio file is stored at row["member"] inside this tar archive. Metadata loading does not automatically decode audio. Individual shards can be several gigabytes. Update REPO_ID if the dataset moves to another namespace.

Limitations

Transcripts are machine-generated weak labels and may contain recognition or normalization errors. Country-level dialect labels are coarse; recordings may contain MSA or code-switching. Web-source coverage and labeling models may introduce demographic, topic, and transcription biases. Use independently transcribed data for evaluation.

Terms of Use

Consistent with the manuscript, this corpus and its derived transcripts and metadata are provided for scientific research and evaluation only. This release does not grant commercial-use permission. Rights in underlying recordings remain with their respective rights holders, and applicable source licenses and platform terms continue to apply. Software recipes are licensed separately.

For questions or removal requests, open a dataset discussion, without publicly disclosing sensitive information.

Citation

Please cite the accompanying manuscript when using this dataset. Publication details will be updated when available.

@misc{liu2026gigaspeech3,
  title = {GigaSpeech 3: Scaling Multi-Dialect Arabic Speech Recognition with Weakly Labeled Data},
  author = {Liu, Junzhe and Tu, Yujie and Wang, Peng and Wang, Haoran and Lin, Guodong and Liao, Leyan and Zhang, Chengquan and Yang, Yifan and Gao, Zhifu and Li, Xiangang and Cheng, Qinyuan and Qiu, Xipeng and Zhang, Weiqiang and Chen, Guoguo and Yu, Kai and Chen, Xie},
  year = {2026},
  howpublished = {Manuscript and dataset},
  url = {https://huggingface.co/datasets/Antares97/GigaSpeech3}
}