Rep. Wexton, confronting degenerative disease, finds her voice through AI

“My new — old — AI voice,” she called it in a recent video introducing the voice to her constituents.

Wexton made headlines this year by using a robotic-sounding speech application to deliver remarks on the House floor. It was a widely hailed display of resilient spirit, but the app didn’t sound like her.

This week, Wexton rolled out her new, more natural-sounding voice as she stood to address the House Appropriations Committee, occasionally touching the walker she’s used since being diagnosed with a Parkinson’s-like condition called progressive supranuclear palsy (PSP).

“For those of you who heard me speak before PSP robbed me of my voice, you may think your ears are deceiving you right now,” she told the committee through an app playing on her iPad. “I assure you they are not. I’m using a new AI model of my voice today — I know, it’s pretty cool.”

The cadence, the tone, the timbre all sounded remarkably like the Wexton who spent five years in the Virginia Senate before being elected to Congress in 2018. She is not seeking reelection this fall because of her health condition.

For her staff, the moment was more than cool.

“It was a big deal. It was something that we didn’t really expect to be able to hear again,” Justin McCartney, Wexton’s communications director, said Friday.

Wexton, 56, has lost none of her wit and intelligence, McCartney said, but has a difficult time communicating verbally because of the neurological disorder. For those who have been close to Wexton over the years, the AI voice “brought back something that a lot of people didn’t realize how much they missed,” he said.

In a written response to questions from The Washington Post, Wexton said Friday that “it will never be ‘me,’ but it’s more me than I or anyone around me ever thought we’d hear again.” After having to turn down speaking engagements and public appearances because she could no longer trust her own voice, Wexton said, the technology has restored her ability “to keep doing my best at this job I love.”

The first sample she heard of her AI voice was a snippet of Hamlet’s “to be or not to be” soliloquy. “My husband got a big smile on his face,” she said. “I haven’t seen him so broadly and genuinely [happy] in too long,” she said, adding that she has enjoyed seeing friends and colleagues react to it.

“Hearing my friend’s voice once again was an incredibly moving moment,” Rep. Abigail Spanberger (D-Va.) said Friday via email. “And while I know this new voice is made possible by a computer, I know that the wise words — and clever but cutting jokes — we’ll hear from it are Jennifer’s and Jennifer’s alone.”

Late last year, when Wexton wanted to address the House in support of a bill aimed at ending Parkinson’s disease, she wrote out remarks and had Rep. Jennifer McClellan (D-Va.) deliver them for her. It was a humbling moment, McClellan said Friday in an interview with The Post.

“For someone so self-reliant who made a career out of using her voice on behalf of other people — first as a prosecutor, then as a legislator — it was very difficult to have to rely on somebody else to speak on your behalf,” McClellan said. “I think the AI software gave her some of her power back in a way that was very moving.”

Wexton acknowledged the pitfalls of AI in her comments to The Post, saying “it’s scary to think about the bad things that someone with bad intentions could do with this technology.” She has limited who on her team has access to the tool, “because using my voice to say something without my consent could cause real problems.”

Experts have increasingly raised the alarm about the potential for the technology to disrupt politics. A report in January from the Wilson Center warned that with national elections scheduled this year in countries around the world, “the risk of ‘blurring the walls of reality,’ as one analyst has put it, through the use of AI-generated and more conventional deepfake productions is disturbingly high.”

The Brennan Center for Justice at New York University School of Law noted late last year that generative AI is already being used in political advertisements and mailers in the United States and is “poised to redefine modern campaigning.” The center’s report noted, though, that Congress has yet to address the issue.

McClellan said Wexton’s adoption of AI emphasizes the need to bring laws and regulations up to date.

The Wexton voice model came about after a video of her using the text-to-speech app in May caught the attention of ElevenLabs, an AI voice start-up based in New York. The company contacted the congresswoman’s staff, and after she gave the green light, staffers spent a couple of weeks compiling more than an hour of audio clips of Wexton speaking before the disease affected her voice.

It took the company just a few days to generate a digital version of Wexton’s voice. A spokesman for ElevenLabs said artificial intelligence makes it possible to not just imitate a voice but modulate tone so it sounds natural, instead of robotic.

“Our model is able to understand the relations between words and to adjust delivery based on context … to produce lifelike, human-sounding speech,” Sam Sklar, a spokesman for ElevenLabs, said via email.

Wexton’s office pays a small subscription fee to ElevenLabs for the service, McCartney said.

ElevenLabs, launched in 2022, has contacted other public figures with a similar offer, such as former Spirit Airlines CEO Ben Baldanza, who has amyotrophic lateral sclerosis, and Greenberg Traurig litigator Lori Cohen, who uses an AI version of her voice to argue cases in the courtroom.

“It’s clear that there’s work to be done to properly protect against potential dangers it poses,” Wexton said in her video. “But it can also provide new, unimaginable and life-changing opportunities for Americans with disabilities.”