CapabilityCognitive automation2022-11-29
Generic cloud speech recognition's word error rate remains relatively high on UK court hearings, researchers found, while a legal-domain model improved on it by 8–9%
Court reporteroccupation page →Event date / reported
2022-11-29
Evidence stage
CapabilityA demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.
Tasks this bears on
Producing and editing transcripts
Turning the record into an accurate written transcript.
Being augmented≈ Platform inference
Where this applies
United Kingdom. Researchers building a transcription tool for the justice sector with an industrial partner say legal transcription is largely met by human transcribers, that the word error rate of a generic cloud speech recogniser remains relatively high on British court rooms because of long hearings, multiple speakers and unusual vocabulary, and that their domain-adapted model outperforms AWS and Whisper with a 9% and 8% word-error improvement; their gold-standard transcripts were made by legal professionals post-editing machine output. It is a workshop preprint with a commercial partner.
What this means
Courtroom audio is hard for generic speech recognition; even the improved systems start from drafts that legal professionals correct.
What it does not yet show
A preprint with a commercial partner and 2022-era models; newer systems may do better.
What you can check
Open the arXiv paper "Better Transcription of UK Supreme Court Hearings" (2211.17094) and find "remains relatively high".
Does it change the assessment?
No — and this stage does not move it either. A "Capability" record is real evidence, but it does not upgrade a task judgement on its own. The 1 linked judgement above stand where they were.
Source
Saadany, Breslin, Orăsan and Walker — Better Transcription of UK Supreme Court Hearings, arXiv:2211.17094 (submitted 29 Nov 2022) · verified 2026-09-30 · Claude (VOLO agent) · interpreted 2026-09-30 · Claude (VOLO agent)
Primary source — published by the party that did this, or the authority of record. No co-signature needed.