Dataset Overview

Dataset Description

Basic Information

• Dataset Name: EdAcc: The Edinburgh International Accents of English Corpus
• Size: 7,542,124,660.365999 bytes
• Download Size: 6,951,164,322 bytes

Structure

• Features:
  • speaker: speaker ID (string)
  • text: transcription of the audio (string)
  • accent: accent labeled by trained linguists (string)
  • raw_accent: self‑reported accent (string)
  • gender: speaker gender (string)
  • l1: speaker's native language, standardized by linguists (string)
  • audio: dictionary containing filename, decoded audio array, and sampling rate (audio)

Splits

• Validation: 9,848 samples, 2,615,574,877.928 bytes
• Test: 9,289 samples, 4,926,549,782.438 bytes

Supported Tasks

• Automatic Speech Recognition (ASR): Models receive audio files and output transcribed text; primary metric is word error rate (WER).
• Audio Classification: Models receive audio files and predict speaker accent or gender; primary metric is percentage accuracy.

Dataset Creation

Data Collection Process

• Participants engaged in relaxed conversations over Zoom and completed detailed questionnaires to collect metadata.
• The questionnaire captured language background, English learning duration, language usage, residence history, relationship with conversation partners, and self‑perceived accent.
• It also gathered demographic information such as age, gender, ethnic background, and education level.
• Conversations were transcribed by professional transcribers, ensuring accurate capture of speech, overlap, background noises, laughter, and hesitations.

License

• License: Public domain, Creative Commons Attribution‑ShareAlike International Public License (CC‑BY‑SA)

Citation

@inproceedings{sanabria23edacc,
   title="{The Edinburgh International Accents of English Corpus: Towards the Democratization of English ASR}",
   author={Sanabria, Ramon and Bogoychev, Nikolay and  Markl, Nina and Carmantini, Andrea and  Klejch, Ondrej and Bell, Peter},
   booktitle={ICASSP 2023},
   year={2023},
}

Contributors

• Thanks to @sanchit-gandhi for adding this dataset.