Millions of words spoken in the U.K.’s highest court risk being misunderstood, misquoted or simply missed because transcribing them accurately is too difficult and too expensive, according to a new study from the University of Surrey.
In a new article published in Applied Sciences, the researchers detail how they built an artificial intelligence system that can automatically transcribe U.K. Supreme Court hearings and link them directly to the written judgments—helping lawyers, academics and the public navigate justice like never before.
Every year, more than 449,000 cases move through U.K. tribunals, yet recordings of court hearings remain hard to use. Traditional transcription is slow, costly and prone to errors. Off-the-shelf speech recognition tools struggle with courtroom language, mishearing “my lady” (pronounced “mee-lady” by barristers when addressing a female judge) as “melody” or legal terms like “inherent vice” as “in your advice.”
To tackle this, researchers developed a custom speech recognition system trained on 139 hours of Supreme Court hearings and legal documents. By fine-tuning the model with specialist vocabulary and court etiquette, the system reduced transcription errors by up to 9% compared with leading commercial tools. It also proved more reliable at capturing crucial entities such as provisions, case names and judicial titles.
Professor Constantin Orăsan, co-author of the study and Professor of Language and Translation Technologies at the University of Surrey, said, “Our courts deal with some of the most important questions in society. Yet the way we record and access those hearings is stuck in the past.
“By tailoring AI to the unique language of British courtrooms, we’ve built a tool that makes justice more transparent and accessible—whether you’re a barrister preparing an appeal or a member of the public trying to understand why a judgment was reached.”
The second part of the project used AI to semantically match paragraphs of judgments with the precise timestamp in the video where the argument was made. A prototype interface now lets users scroll through a judgment, click on a paragraph and instantly watch the relevant exchange from the hearing. Tests showed the system correctly linked text and video with an F1 score of 0.85.
An F1 score is a way of measuring how well a system balances two things:
- Precision—of all the results it gave, how many were actually correct.
- Recall—of all the correct results that existed, how many it managed to find.
It punishes a system that is very good at one but bad at the other. It ranges from 0 to 1:
- 1.0 means perfect precision and recall (the system found everything and made no mistakes).
- 0 means total failure.
Evaluation with real users showed that their productivity is dramatically increased when using the UI. Without AI assistance, a legal expert needed 15 hours to identify 10 links, whereas with AI support they were able to validate 220 links in just three hours.
The tool is already attracting interest from legal bodies, including the U.K. Supreme Court and the National Archives. By reducing hours of manual searching into seconds, it promises to help lawyers prepare cases, speed up legal training and allow the public to see how decisions are formed.
More information:
Hadeel Saadany et al, Employing AI for Better Access to Justice: An Automatic Text-to-Video Linking Tool for UK Supreme Court Hearings, Applied Sciences (2025). DOI: 10.3390/app15169205
University of Surrey
Citation:
AI transcribes UK Supreme Court hearings and links them to written judgments (2025, September 16)
retrieved 16 September 2025
from https://techxplore.com/news/2025-09-ai-uk-supreme-court-links.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.
