{"success":true,"data":{"author":"Ramon Sanabria, Nikolay Bogoychev, Nina Markl, Andrea Carmantini, Ondrej Klejch, Peter Bell (University of Edinburgh)","bucketAccessType":"public","bucketName":"edinburghcstr/edacc","bucketPrefix":"","category":"Speech & Audio","cloudProvider":"huggingface","createdAt":"2026-11-14T14:15:17.888Z","description":"The Edinburgh International Accents of English Corpus (EdAcc) is an automatic speech recognition dataset of 40 hours of English dyadic conversations between speakers with a diverse set of accents. It includes a wide range of first- and second-language varieties of English and a linguistic background profile for each speaker. Its authors show that current public and commercial ASR models, which perform well on standard benchmarks, degrade substantially on these speakers, so EdAcc is used to measure accent robustness and fairness of speech recognition systems. It also supports accent classification. A public leaderboard ranks models by word error rate on the dev and test sets. This repository, published by the University of Edinburgh's Centre for Speech Technology Research, holds the dev and test sets as 16 Parquet shards under data/, with audio, transcript, speaker and accent metadata (measured 2026-11-14; the Hub reports about 36 GB). Load with datasets.load_dataset('edinburghcstr/edacc'). Released under CC BY-SA 4.0; cite Sanabria et al., ICASSP 2023.","displayTitle":"EdAcc: Edinburgh International Accents of English Corpus","id":"3737b67d-ce1e-5671-ae86-93d476cf33d5","license":"CC-BY-SA-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/edinburghcstr/edacc","objectCountEstimate":18,"probeStatus":"ok","referenceUrl":"https://arxiv.org/abs/2303.18110","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"edinburghcstr/edacc\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"edinburghcstr/edacc\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=edinburghcstr, size_categories:10K<n<100K, format:parquet, downloads=6627"],"format":"mixed","generatedAt":"2026-11-14T14:15:30.805Z","layout":{"example":"data/test-00000-of-00010-f0aceb1ca4406ff1.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"data/test-00000-of-00010-f0aceb1ca4406ff1.parquet","truncatedAt":512}],"sampleObjects":["data/test-00000-of-00010-f0aceb1ca4406ff1.parquet","data/test-00001-of-00010-856b016d9d438ff3.parquet","data/test-00002-of-00010-2b021baedb4deb8a.parquet","data/test-00003-of-00010-4de275e704375a02.parquet","data/test-00004-of-00010-806407c9bc68112a.parquet"],"sidecars":[{"bytes":"---\nconfigs:\n- config_name: default\n  data_files:\n  - split: validation\n    path: data/validation-*\n  - split: test\n    path: data/test-*\ndataset_info:\n  features:\n  - name: speaker\n    dtype: string\n  - name: text\n    dtype: string\n  - name: accent\n    dtype: string\n  - name: raw_accent\n    dtype: string\n  - name: gender\n    dtype: string\n  - name: l1\n    dtype: string\n  - name: audio\n    dtype: audio\n  splits:\n  - name: validation\n    num_bytes: 2615574877.928\n    num_examples: 9848\n  - name: test\n    num_bytes: 4926549782.438\n    num_examples: 9289\n  download_size: 6951164322\n  dataset_size: 7542124660.365999\ntask_categories:\n  - automatic-speech-recognition\n  - audio-classification\n---\n\n## Dataset Description\n\n- **Homepage:** [EdAcc: The Edinburgh International Accents of English Corpus](https://groups.inf.ed.ac.uk/edacc/index.html)\n- **Paper:** [The Edinburgh International Accents of English Corpus: Towards the Democratization of English ASR](https://arxiv.org/abs/2303.18110)\n- **Leaderboard:** [EdAcc Leaderboard](https://groups.inf.ed.ac.uk/edacc/leaderboard.html)\n\n\n# EdAcc: The Edinburgh International Accents of English Corpus\n\nThe Edinburgh International Accents of English Corpus (EdAcc) is a new automatic speech recognition (ASR) dataset \ncomposed of 40 hours of English dyadic conversations between speakers with a diverse set of accents. EdAcc includes a \nwide range of first and second-language varieties of English and a linguistic background profile of each speaker. \nResults on latest public, and commercial models show that EdAcc highlights shortcomings of current English ASR models,\nwhich perform well on existing benchmarks, but degrade significantly on speakers with different accents.\n\n## Supported Tasks and Leaderboards\n\n- Automatic Speech Recognition (ASR): the model is presented with an audio file and asked to transcribe the audio file to written text. The most common evaluation metric is the word error rate (WER). The task has an active leaderboard which can be found at https://groups.inf.ed.ac.uk/edacc/leaderboard.html and ranks models based on their WER scores on the dev and test sets\n- Audio Classification: the model is presented with an audio file and asked to predict the accent or gender of the speaker. The most common evaluation metric is the percentage accuracy.\n\n## How to use\n\nThe `datasets` library allows you to load and pre-process EdAcc in just 2 lines of code. The dataset can be \ndownloaded from the Hugging Face Hub and pre-processed by using the `load_dataset` function. \n\nFor example, the following code cell loads and pre-processes the EdAcc dataset, and subsequently returns the first sample\nin the validation (dev) set:\n\n```python\nfrom datasets import load_dataset\n\nedacc = load_dataset(\"edinburghcstr/edacc\")\nsample = edacc[\"validation\"][0]\n```\n\nUsing the datasets library, you can also stream the dataset on-the-fly by adding a `streaming=True` argument to the \n`load_dataset` function call. Loading a dataset in streaming mode loads individual samples of the dataset at a time, rather \nthan downloading the entire dataset to disk. The only change is that you can no longer access individual samples using \nPython indexing (i.e. `edacc[\"validation\"][0]`). Instead, you have to iterate over the dataset, using a for loop for example:\n\n```python\nfrom datasets import load_dataset\n\nedacc = load_dataset(\"edinburghcstr/edacc\", streaming=True)\nsample = next(iter(edacc[\"validation\"]))\n```\n\nFor more information, refer to the blog post [A Complete Guide to Audio Datasets](https://huggingface.co/blog/audio-datasets#streaming-mode-the-silver-bullet).\n\n## Dataset Structure\n\n### Data Instances\n\nA typical data point comprises the loaded audio sample, usually called `audio` and its transcription, called `text`. \nSome additional information about the speaker's gender, accent and native language (L1) are also provided:\n\n```\n{'speaker': 'EDACC-C06-A',\n 'text': 'C ELEVEN DASH P ONE',\n 'accent': 'Southern British English',\n 'raw_accent': 'English',\n 'gender': 'male',\n 'l1': 'Southern British English',\n 'audio': {'path': 'E","encoding":"utf8","key":"README.md","name":"README.md","size":6947,"truncated":true}],"sourceObjectKey":"data/test-00000-of-00010-f0aceb1ca4406ff1.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"edinburghcstr/edacc\")\nprint(ds)","starterSamplePath":"data/test-00000-of-00010-f0aceb1ca4406ff1.parquet","totalObjects":18},"summary":"The Edinburgh International Accents of English Corpus (EdAcc) is an automatic speech recognition dataset of 40 hours of English dyadic conversations between speakers with a…","title":"EdAcc: Edinburgh International Accents of English Corpus","updatedAt":"2026-11-14T17:47:46.661Z","visibility":"public","websiteUrl":"https://groups.inf.ed.ac.uk/edacc/index.html"}}