zhifeixie/Voices-in-the-Wild-2M

Voices in the Wild

Project Page | Paper | GitHub

Voices in the Wild (Voices-in-the-Wild-2M) is a large-scale automatic speech recognition (ASR) dataset designed for robustness training and evaluation under diverse, real-world acoustic conditions. It covers 7 classic acoustic phenomena (including noise, far-field speech, obstruction, echo/reverberation, recording artifacts, electronic distortion, and transmission dropout) and 54 physically plausible compound scenarios.

The dataset was introduced as part of the Mega-ASR framework to address the "acoustic robustness bottleneck" where models produce omissions or hallucinations under severe compositional distortions.

Data Fields

• file_name: relative path to the audio file.
• audio_path: audio path retained for local tooling.
• text: transcription alias copied from answer.
• answer: reference transcription.
• question: transcription instruction.
• subset: normalized acoustic condition category.
• prediction: empty placeholder for model output.
• name: public sample identifier.
• index: integer sample index.

Dataset Size

• Total examples: 645,925
• Subset categories: 54

Loading

from datasets import load_dataset, Audio

# Load the dataset from the Hub
ds = load_dataset("zhifeixie/Voices-in-the-Wild-2M")
ds = ds.cast_column("audio", Audio())

# Access an example
print(ds["far_field"][0])

Citation

@misc{xie2026megaasrinthewild2speechrecognition,
      title={Mega-ASR: Towards In-the-wild^2 Speech Recognition via Scaling up Real-world Acoustic Simulation},
      author={Zhifei Xie and Kaiyu Pang and Haobin Zhang and Deheng Ye and Xiaobin Hu and Shuicheng Yan and Chunyan Miao},
      year={2026},
      eprint={2605.19833},
      archivePrefix={arXiv},
      primaryClass={cs.SD},
      url={https://arxiv.org/abs/2605.19833},
}