What’s the Future of Court Reporting? (Part 2)

The Ninth Judicial Circuit Court of Florida has both real-time court reporters and digital reporters on staff. The digital reporters remotely oversee several courtrooms at once where audio is being recorded. While doing so, they make annotations — such as indicating who is speaking when — that are tagged in the audio file to help when they go back later to write out an official transcript, said trial court administrator Matt Benefiel.

But Benefiel’s court still struggles to hire enough real-time and digital reporters, worrying about what happens when those on staff retire or depart.

The court is currently testing a real-time speech-to-text transcription solution from the company For The Record at its juvenile courthouse. Benefiel said the tool could help digital reporters cover more cases and produce records faster, providing them with rough transcripts.

Real-time stenographic court reporters are a trickier problem, and these hiring struggles are particularly acute.

“The supply of real-time stenographers is just dwindling. We have the staff but there’s no one to replace them when they leave,” Benefiel said.

But while the technology has been advancing, he said the tools aren’t quite there yet.

“Right now, it’s a race before we don’t have court reporters,” Benefiel said.

CLARITY AND PRIVACY

Sue Terry leads the National Court Reporters Association’s NCRA STRONG task force, which advocates against using electronic recording and automatic speech recognition (ASR) in place of court reporters. Terry said digital recordings could serve as backups but are risky as the main record.

Risks include malfunctions that leave no record or significant gaps. Law360 gives the example of a 2018 glitch that left a court without a recording for the first day of a trial, and a 2019 case where the recording system struggled to clearly capture speech, resulting in a transcript with 108 passages listed inaudible. Issues might stem from tech errors, user mistakes or implementation problems.

Courtrooms tend to have background noise, multiple speakers and people talking over each other, which are challenging for audio recordings, Terry said. A reporter present during the hearing, meanwhile, could simply ask participants to repeat themselves.

Benefiel said his court’s acoustic-friendly design has helped, with low ceilings, carpet and paneling rather than hardwood floors or marble walls. Such systems also should have separate audio channels for different speakers, and microphones spaced out to better avoid catching other speakers.

For The Record CEO Tony Douglass said his offerings are designed so each mic prioritizes the loudest audio, helping it find the person speaking into it.

Capturing each speaker’s audio separately lets speech-to-text solutions transcribe speakers simultaneously, Douglass said — whereas a stenographic reporter can only transcribe one person then would have to ask for repetition.

Terry said privacy is a concern with ASR. If an attorney forgets to mute a mic while talking to a client, the system could enter it into the official public record, Terry said. Benefiel’s court relies on digital reporters overseeing the final transcripts to edit out private sidebars.

For Benefiel’s court, another feature is key to using real-time speech-to-text more widely: the ability to edit annotations in the transcript, especially speakers’ names. That’s important in juvenile matters, where five different family members might share the same defendant mic, making it difficult to separate who’s who. Digital reporters could add clarifications into the automated transcript, like naming who’s speaking into a shared mic.

If the court gets that capability from the vendor and it works well for juvenile cases, it’ll expand use to felony, misdemeanor and perhaps even capital cases.

ACCURACY

Litigants should have the right to know if ASR was involved in drafting their transcript, as well as which model was used and how it was tested for accuracy, Terry said. That information could be important to appeals.

There’s a risk speech-to-text tools will mix up word boundaries, potentially transcribing something like “below knee amputation” as “baloney amputation” during a medical malpractice case, she said. The tools don’t always perform well across genders and races, either, and sometimes they mistake the sound of typing for words.

“While automatic speech recognition is coming along for some disciplines, it certainly isn’t to the level where our justice system can depend on it, certainly not without human intervention,” Terry said,

Studies have found that both humans and tech trip up.

Stanford University researchers testing ASR from companies like Apple, Amazon and Google in 2019 found significantly higher error rates transcribing Black speakers and African American Vernacular English (AAVE) compared to white speakers. The systems got at least half of the words wrong in over 20 percent of speech samples from Black speakers compared to less than 2 percent of samples from white speakers.

Meanwhile, a 2019 study of 27 Philadelphia court reporters found many struggled with AAVE. Professionals had to transcribe with 95 percent accuracy at various word-per-minute speeds to get certified, but those tests focused on Mainstream American English. When hearing AAVE, the court reporters failed to clear that bar, and 31 percent of the errors changed key details like the who, what, when or where.

Douglass said systems that pair recorded audio and written text act as a safeguard against human or software mistakes, allowing for double-checking a disputed or unclear part of the transcript: “You’ve got access to the recording, and you can find it instantly and play it back and verify for yourself what was actually said.”

THE FUTURE

If the technology makes the same leaps in the next five years that it did in the past five, it might reach a point where speech-to-text systems don’t need to be overseen by reporters but only by technologists to make sure it’s working well, Benefiel predicted.

Discussing the ASR tech landscape — not a particular product — Terry said the tools do not beat real-time stenographic reporters on speed, while they lack discretion to avoid transcribing private conversations, among other issues.

“Those are things that I don’t think will be replaced in our lifetime: You can’t replace the human brain,” Terry said. “Automatic speech recognition, at this point, even if it were accurate, doesn’t offer anything we don’t offer now. … If we saw a benefit to it [ASR], we’d be trying to incorporate it, for sure.”

Douglass said technologies are opening the door to re-envisioning what a court record should look like, and it’s open to discussion whether reliable tech is better than a single human’s interpretation of what they heard.

Not all jurisdictions tightly define what an official court record must look like, leaving some judges to decide whether an AI-produced transcript could serve. Douglass said they’re working now with a judge who might make it the official record in a case-by-case basis, which could be a glimpse at the future.

This post was originally published on this site