Open data for the UK accent benchmark

Scores are only worth something if you can rerun them. This is what we publish, and what we only point to.

What we publish

We will publish these with every round, under our own name and a dated version.

Scores. Every row, with the model version, region and run date.
Clip lists. The exact files used, pinned to a dataset version.
Scripts. The phone-line and noise steps, the normaliser settings and the scoring code.
Vendor replies. Printed beside the results they answer.

What we point to

We do not rehost other people’s recordings.

Share-alike audio stays where its owners put it. We link to the original, pin the version and publish the script that rebuilds our phone-line copies on your machine.

The speakers are real people, so we never try to work out who they are.

Sources and licences

Read from each dataset’s own page, 26 October 2026.

Datasets, their licences and what we share
Dataset	Licence	What we share
OpenSLR 83, UK and Ireland English	CC BY-SA 4.0	Clip list and rebuild script
Edinburgh International Accents of English	CC BY-SA 4.0	Clip list and rebuild script
Cekura turn-taking benchmark	CC BY 4.0	Linked, with credit

Licences change. We re-read each source every quarter and date any change. The method

Run a test on your own agent.

Get a free report Book a call