Late in July, Rep. Jennifer Wexton did what she has done innumerable times since being elected to Congress five years ago: she took her place at the podium on the House floor and spoke to her colleagues. But this time, for the very first time, her voice was created solely by AI.
Last year, the Virginia Democrat received the devastating news that she suffers from progressive supranuclear palsy (PSP), an extremely uncommon and potentially fatal brain condition. She made the announcement that she will not run for reelection when her current term ends this year after receiving the diagnosis. She is now unable to walk normally and has a weaker, less clear voice as a result of the illness. However, Wexton was able to reproduce it using vintage recordings with the assistance of ElevenLabs.
Since that same technology had previously created a deepfake of President Joe Biden’s voice in January, that House floor moment brought assistive technologies into the national limelight and offered a positive contrast to the dread typically linked with AI. As Congress considers whether to regulate AI, Wexton and her colleagues must now balance the pros and cons of the technology.
Using her AI-generated voice, Wexton explained to the HEADLINESFOREVER Tech podcast why she has become such an advocate for assistive technology during her last term in government and how that discussion now has much more personal resonance. There must be a positive aspect to this illness, she said.
The following has been condensed and made clearer. The complete interview with Wexton is available for listening online at:
Commonplace Congressional duties, such as addressing on the floor, interrogating witnesses in committee, and conducting interviews like this, became impossible due to PSP’s effect on the loudness and clarity of my voice. For a while, I was even forced to decline chances to speak publicly, which was incredibly annoying.
I accepted ElevenLabs’ offer to have an artificial intelligence model of my voice created after a while of battling with the robotic text-to-speech program to make it seem more natural in terms of pronunciation, rhythm, and tone. It just took a few days after my team transmitted over an hour of audio recordings of me giving speeches or other public statements for the AI model to be ready. It is quite fantastic to have a new “old” AI voice for myself.
We finished the AI voice model in early July as a team, and I got it the day before I was supposed to attend President Biden’s signing of the National Plan to End Parkinson’s Act into law at a small ceremony in the Oval Office.
Following my diagnosis, I made it a point to advocate for this new bill in Congress. I met with senators and representatives from both parties and shared my story of how difficult it was to get a diagnosis and how I’ve struggled to find symptom management treatments. I also talked about the impact this bill could have on our fight against Parkinson’s and related diseases like PSP, and how much more funding we could receive as a result. After a lot of hard work behind the scenes, it passed both houses with unanimous support from both parties, and I couldn’t be prouder.
It was an incredible experience for my family and me to be at the White House to witness the signing of this historic bill into law. I wished I could have made it even more memorable by introducing the president to my brand new artificial intelligence voice. Legislator Gus Bilirakis, with whom I work closely and who is also one of the bill’s co-leads, had never heard me speak before the PSP. I thought my mom was going to burst into tears, but she was so strong. Witnessing the impact of my advocacy allowed me to express my deepest gratitude to President Biden and my family. I will speak in my own voice, not a robotic one.
When you first heard an AI-generated version of your voice, how did it feel?
I initially heard a sample of my artificial intelligence voice doing a soliloquy from Hamlet while my husband was here. The question is, “To be or not to be.” Consequently, we were both enthralled by it the moment we heard it. I wept with joy. For one, it sounded remarkably more natural than the text-to-speech tool I had been utilizing, and secondly, it sounded just like me. In the middle of a sentence, my AI voice paused to catch its breath. It was fantastic. My spouse’s face lit up with joy. It had been far too long since I last saw him with such a wide, sincere smile. Friends and coworkers who hadn’t heard from me in a while texted me numerous times to express their loneliness.
Although my AI voice will never sound exactly like me, it is more like me than I had anticipated. Not only has it benefited my professional life, but it has also encouraged me to continue working hard at the job I enjoy. It’s meaningful to me because it gave me a voice, and I intend to utilize it to speak up for others who are dealing with the same kinds of health and ability issues I am. I have been able to open up about my background, struggles, and battle for change in Congress.
Our ElevenLabs-created model is suitable for usage in formal speeches and similar occasions. I can change the model’s attributes. It takes only seconds to use the AI model through a regular internet browser interface to make the audio after my team and I have worked together on speeches in the same way as before. Creating additional options for various speaking styles is my next goal with this AI voice model. For instance, the present model may come across as somewhat stuffy and unnatural at times. Everything has the potential to sound like a grand declaration. Therefore, I will not use it to politely request that my spouse bring me the ketchup. However, there are a plethora of audio recordings of me in various contexts, such as television interviews or campaign events, due to the more than twenty years I’ve spent serving in public office. I’m interested in constructing a model using those movies so I can enhance my AI voice for different situations and be more dynamic.