RIVERVIEW, PORTLAND (AP) — Before last summer, Alexis “Lexi” Bogan had a vibrant voice.

In the car, she enjoyed belting out Zach Bryan and Taylor Swift songs. She never stopped laughing, whether she was dealing with misbehaving toddlers or having a political debate with friends around a fire pit in the backyard. She sang soprano in the chorus in high school.

That voice vanished after that.

A life-threatening tumor that was located close to the back of her brain was removed by doctors in August. A month after the breathing tube was removed, Bogan struggled to speak and swallow, making it difficult for her to greet her parents. Although months of therapy aided in her recuperation, she still has speech impairments. Her own family members, friends, and strangers find it difficult to comprehend what she is attempting to say.

The 21-year-old’s voice returned in April. She can call upon an artificial intelligence-generated voice clone, which is not the actual one, through a mobile application. Her artificial intelligence voice, which sounds remarkably lifelike, was trained on a 15-second clip of her adolescent voice, taken from a cooking demonstration video she took for a high school assignment. She can now say nearly anything she wants.

The app reads aloud everything she writes onto her phone in a matter of seconds.

At a Starbucks drive-thru, Bogan’s AI voice said, “Hello, can I please get a grande iced brown sugar oat milk shaken espresso?” as she held the phone out her car window.

Experts have issued a warning that the quickly advancing AI voice-cloning technology might increase the impact of phone scams, tamper with democratic elections, and violate the dignity of those who, whether alive or dead, never gave their permission to have their voices recorded to say things they never said.

It’s being utilized to generate deepfake robocalls that imitate President Joe Biden to voters in New Hampshire. Authorities in Maryland have accused a high school athletic director of fabricating an audio clip purporting to show the principal of the school making racist remarks using artificial intelligence (AI).

But Bogan and a group of medical professionals at the Lifespan hospital group in Rhode Island think they’ve discovered a use that outweighs the dangers. With OpenAI’s innovative Voice Engine, Bogan is among the first persons — and the only one with her condition — to successfully restore a lost voice. ElevenLabs, a startup, and other AI providers have explored comparable technologies for individuals with speech impairments and loss. One such individual is a lawyer who now utilizes her voice clone in court.

As the technology advances, Dr. Rohaid Ali, a neurosurgery resident at Rhode Island Hospital and the medical school at Brown University, said, “We’re hoping Lexi’s a trailblazer.” He said that millions of patients with crippling strokes, throat cancer, or neurodegenerative illnesses would gain from it.

Another resident working on the pilot, Dr. Fatima Mirza, stated, “We should be conscious of the risks, but we can’t forget about the patient and the social good.” “With our assistance, Lexi can regain her authentic voice and communicate in ways that are most authentic to her.”

The married couple Mirza and Ali attracted the interest of OpenAI, the company that makes ChatGPT, with a prior study at Lifespan that used an AI chatbot to streamline patient medical permission forms. When searching for potential medical uses for its new AI speech generator early this year, the San Francisco-based startup made contact.

Bogan was still healing after surgery very slowly. Headaches, blurred eyesight, and a sagging face were the first signs of the sickness last summer, which alarmed the doctors at Hasbro Children’s Hospital in Providence. They found a golf-ball-sized vascular tumor tangled in cranial nerves and blood vessels, pressing on her brain stem.

Dr. Konstantina Svokos, a pediatric neurosurgeon, described the process of removing the tumor and controlling the bleeding as a struggle.

Bogan’s capacity to eat and speak was hindered by the placement and severity of the tumor as well as the difficulty of the 10-hour surgery, which damaged her vocal cords and tongue muscles, according to Svokos.

“It seems as though a portion of my identity was stolen when I went silent,” Bogan remarked.

This year saw the release of feeding tubes. She is still receiving speech treatment, which helps her talk clearly in calm environments, but there is no indication that she will regain the full clarity of her original voice.

“I was beginning to lose track of my voice at some point,” Bogan remarked. “I’ve been so accustomed to my current voice now.”

She would pass the phone to her mother to answer calls whenever the family’s North Smithfield, Rhode Island, house rang. Every time they went to a noisy restaurant, she felt like she was bothering her pals. Her dad had trouble understanding her because he is hard of hearing.

Doctors were searching for a pilot patient to test OpenAI’s technologies on back at the hospital.

“Lexi was the first person that Dr. Svokos thought of,” Ali stated. We didn’t know what to expect from Lexi when we asked whether she would be interested. She was willing to give it a shot and see how it performed.

In order to “train” the AI system on her speech pattern, Bogan had to go back a few years and locate a suitable recording of her voice. She was demonstrating how to prepare spaghetti salad in the video.

Her physicians gave the AI system only a 15-second clip on purpose. Certain parts of the video are ruined by the noises of cooking. Additionally, it was all that OpenAI required, which was a gain over earlier systems that required many longer samples.

Additionally, they were aware that extracting value from 15 seconds could be crucial for prospective patients whose voice is lost to the internet. Maybe just a quick voicemail to a relative will do.

Everyone was astounded by the voice clone’s quality when they tested it for the first time. A few occasional errors, like a mispronounced word or a missing tone, were largely undetectable. Doctors gave Bogan a unique phone app in April that she is the only one who can use.

Her mother Pamela Bogan, with tears in her eyes, remarked, “I get so emotional every time I hear her voice.”

Lexi Bogan continued, “It’s great that I can have that sound again,” adding that it has “boosted my confidence to a point where it was somewhat before all of this happened.”

She now submits comments through the app, maybe forty times a day, hoping that it would assist other patients. Speaking with the children at the preschool where she works as a teaching assistant was one of her initial trials. She entered “ha ha ha ha,” anticipating a mechanized reply. She was surprised to hear that it was her old chuckle.

She has asked where to find things at Marshall’s and Target using it. It has enabled her and her father to reunite. Additionally, it has made ordering fast food easier for her.

With the goal of introducing the technology to hospitals worldwide, Bogan’s medical team has begun cloning the voices of other consenting patients from Rhode Island. OpenAI stated that it is taking cautious steps to increase the usage of Voice Engine, which is not yet accessible to the general public.

A number of smaller AI businesses already provide voice-cloning services more broadly or offer them to Hollywood studios. While the majority of voice-generation suppliers claim to forbid impersonation or misuse, their methods of enforcing their conditions of use differ.

According to Jeff Harris, OpenAI’s principal developer for the program, “we want to make sure that everyone whose voice is used in the service is consenting on an ongoing basis.” “We wish to ensure that political contexts do not employ it. As a result, we’ve adopted a strategy of allowing very specific people access to the technology.

According to Harris, the next task for OpenAI is to create a safe “voice authentication” application that would allow users to mimic just their own speech. For a child like Lexi, who experienced an abrupt loss of her verbal abilities, that might be “limiting,” he said. Therefore, we do believe that in order to grant a little bit more unrestricted access to the technology, we’ll need to establish high-trust connections, particularly with medical providers.

Bogan’s focus on considering how the device could benefit others with comparable or more severe speech impairments has impressed her doctors.

“She has thought of ways to tweak and change this throughout the entire process,” Mirza remarked. “She has greatly inspired us,” the group said.

Bogan envisions an AI voice engine that transcends previous treatments for speech recovery, such as the robotic-sounding electrolarynx or a voice prosthesis, by integrating with the human body or translating words in real time. For the time being, she must tinker with her phone to get the voice engine to speak.

As she gets older, she becomes less certain of what will happen, and her AI voice still sounds like it did when she was a youngster. She suggested that the technology might “age” her AI voice.

For the time being, she stated, “I have something that helps me find my voice again, even though I don’t have my voice fully back.”