Industries / Finance and business services / Legal and professional services / Court reporting
Voice writing and digital recording
Making the record by repeating every word into a masked microphone, or by recording the courtroom on audio or video.
7
Milestones achieved, from 1940s to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1940s
Speaking into a mask
Horace Webb developed the Stenomask, a mask with a microphone inside, into which a reporter repeats every word said in the room without being heard. The National Verbatim Reporters Association (NVRA), founded in 1967, says the method "was born in a Chicago courtroom in the early forties"; a 2005 AP report says the US Navy chose it during World War II and calls Webb a Washington, D.C. shorthand reporter. US federal courts list the stenomask among the accepted ways to record proceedings verbatim.
Credit
Horace Webb, with Harold Steinman and Frank Kenny
1981
Video as the trial record
From late 1981 Judge James S. Chenault had every trial in his courtroom in Richmond, Kentucky, recorded on videotape, with no stenographer. A system in Jefferson County began in February 1985. A 1985 evaluation by the National Center for State Courts found the video record cost about the same as other methods, "but its benefits are far greater."
Credit
Judge James S. Chenault; Chief Judge Laurence E. Higgins; Jefferson Audio Video Systems (JAVS)
1982
Federal courts may record
The Federal Courts Improvement Act of 1982 let US district courts make the record by electronic sound recording "or any other method". A 1965 law had allowed recording only as a backup to the shorthand reporter. A 1983 Federal Judicial Center test in 12 courtrooms found that audio, well managed, gave an accurate record at lower cost: $18,604 a year per courtroom, against $40,514 for a stenographic reporter, by the Center's figures.
LessonTwo operators in the test reported equipment failures, one lasting half a day and one on five separate days, and no backup recorders had been bought.
Credit
US Congress; J. Michael Greenwood, Julie Horney, M. Daniel Jacoubovitch, Frances B. Lowenstein and Russell R. Wheeler (Federal Judicial Center)
1997
Voice writers use speech recognition
Voice writers began sending their masked speech to speech recognition software trained on their own voice, which types it out as they talk. Linda Drake, later NVRA president, said she bought her system in 1997. An AP report in 2005 put the number of US voice writers at as many as 15,000.
Credit
Linda Drake; NVRA
2000
The video is the record
Spain's civil procedure law of 2000 says oral hearings "se registrarán" (shall be recorded) in sound and image, and the current text says recorded hearings "no podrán transcribirse" (may not be transcribed) unless a law requires it. Spain's state prosecutor's office dates the duty to record from 1 October 2010, under a 2009 law. Brazil's National Council of Justice ruled in 2010 that testimony recorded on video needs no transcript, and Brazil's 2015 civil procedure code lets a whole hearing be recorded.
Credit
Cortes Generales of Spain; Conselho Nacional de Justiça of Brazil
2012
Crown Courts switch to audio
England and Wales piloted DARTS (Digital Audio Recording, Transcription and Storage) in seven Crown Courts from October 2009 and rolled it out nationwide in June 2011. The last Crown Court stenographers stopped work in March 2012, the Solicitors Journal reported. HM Courts and Tribunals Service (HMCTS) says it must "record and store audio from all Crown Court hearings."
LessonLawyers warned at the switch about witnesses "whose voices are so low they are not picked up.".
Credit
HM Courts and Tribunals Service
2026
A record for poorer litigants
In September 2024 the Los Angeles Superior Court let judges use electronic recording in some family, probate and civil cases beyond what state law allowed, over the opposition of the county's court reporters' association. On 10 August 2026 the California Supreme Court held that a court must give a litigant who cannot pay "some way" to get an official verbatim record, "including by electronic recording if an official reporter is unavailable." NCRA's journal noted that the ruling "did not grant courts unlimited authority to use ER".
Credit
Presiding Judge Samantha P. Jessner; Chief Justice Patricia Guerrero, for the California Supreme Court in Family Violence Appellate Project v. Superior Court
Watch for
Recordings as Alabama's record
The Supreme Court of Alabama ordered a pilot in which audio recordings "shall serve as the official recording of the proceeding and as the source record from which transcripts are prepared", with yearly reports to the court on results "including accuracy, cost".
Achieved whenthe first yearly report is published with accuracy and cost figures, preferably checked by someone outside the courts' own administration.
Claimed by
Supreme Court of Alabama, June 2026
Speech recognition inside voice software
Stenograph and Speechmatics said Speechmatics' speech recognition now runs inside CATalyst VP, Stenograph's software for voice writers, entirely on the reporter's own computer, with Dragon still available as an option. It will also power CheckIt Plus, an add-on used by stenographers and voice writers "to deliver cleaner realtime and rough drafts while reducing overall editing time." The release gives no figures.
Achieved whena court, school or test lab publishes accuracy or editing-time measurements with and without it.
Claimed by
Stenograph and Speechmatics, July 2026
Open
01
Flag bad audio live
A monitor for courtroom recording that warns the clerk during the hearing when a microphone channel goes silent, when speech is too quiet to transcribe, or when two people talk over each other, so the judge can ask for a repeat.
The briefCredit · unclaimed02
Labelled courtroom recording faults
A public set of courtroom recordings in which two people, working separately, have marked every overlap, inaudible passage and dead channel, with their agreement reported, so that monitors and transcription tools can be compared on the same audio.
The briefCredit · unclaimed
Open 01 · the brief
Flag bad audio live
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A monitor for courtroom recording that warns the clerk during the hearing when a microphone channel goes silent, when speech is too quiet to transcribe, or when two people talk over each other, so the judge can ask for a repeat.
Why it is still open
For The Record's monitoring warns of "disconnected microphones, lost audio sources, or storage concerns", and JAVS microphones change light colour when they are not recording. Both are equipment alarms. I found no product or paper that warns about overlapping or too-quiet speech as it happens. The cost of missing it shows in Ohio, where an appeals court ordered a new trial in 2024 because the trial's audio was in parts "almost completely inaudible".
Built with what exists
pyannote's open-source overlapped-speech detection model (MIT licence); a voice-activity detector such as Silero VAD and a loudness meter; hearing videos and transcripts that the Post Office Horizon IT Inquiry publishes after each hearing.
First step
Download five hearing days, mark overlaps and inaudible stretches by hand, and run pyannote's overlap detection with a simple loudness threshold.
Done looks like
On at least 10 hours of inquiry audio with problem stretches marked by hand: the share of marked problems caught, the false alarms per hour, and warnings within about 2 seconds, published with the code and the labels.
Open 02 · the brief
Labelled courtroom recording faults
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of courtroom recordings in which two people, working separately, have marked every overlap, inaudible passage and dead channel, with their agreement reported, so that monitors and transcription tools can be compared on the same audio.
Why it is still open
The open overlap detection model I found was trained on meeting recordings, and the courtroom audio work I found, a 2021 paper on telling apart speakers in US appeals court arguments, labels who spoke when, not recording faults. I found no labelled set of courtroom recording faults.
Built with what exists
public hearing recordings whose terms allow reuse, such as UK public inquiry hearings and US federal appeals court audio; ELAN or Label Studio for marking audio; pyannote to suggest stretches for people to check.
First step
Check the reuse terms of two sources of hearing audio and label one hour from each.
Done looks like
At least 20 hours labelled, the agreement between the two labellers reported, and the set released under terms that match its sources.