AI makes justice more transparent – researchers (www.bbc.co.uk)

Researchers at the University of Surrey have developed a custom speech-recognition system trained on 139 hours of Supreme Court hearings and associated legal documents to produce more accurate transcripts of British court proceedings. The domain‑specific automatic speech recognition (ASR) model reportedly cuts transcription errors by up to 9% versus leading commercial tools by learning the “unique language of British courtrooms,” including legal terminology, speaker turns and courtroom conventions. The team says the result makes courtroom output easier to access and understand, helping make justice more transparent and usable for the public and practitioners alike. The project’s second component uses semantic matching to link paragraphs of written judgments to the exact video timestamps where arguments occurred, creating precise, searchable cross‑references between spoken submissions and final rulings. That feature has attracted interest from the UK Supreme Court and the National Archives and could speed legal research, improve archival quality, and enable more robust public scrutiny of judicial processes. Key caveats are the project's relatively modest training corpus and domain specificity, but the approach demonstrates how tailored AI systems can meaningfully improve legal documentation, searchability, and accountability in court systems.