How 8-second scratch audio from VHS tape gave her mother her voice

“After spending such a long time, I really didn't remember my voice,” says Sarah Ezekiel, who visits the BBC. “When I heard it again, I felt like I was crying. It's kind of a miracle.”

Motor neuronal disease (MND) developed when she left Sarah at the age of 34 without using her voice and hands.

As they grow up, her children Aviva and Eric have heard her speak through a machine with an emotionally-free robotic voice.

But 25 years later, Artificial Intelligence (AI) recreated Mama's true voice from just eight seconds of audio on Scratch's VHS tape.

Sarah speaks to the BBC with her eye gaze technology. I use my camera to track my eyes and look at the letter on the screen in front of me.

The sound of her young voice blew through.

“Miracle” Sarah began when Smartbox, a Bristol-based assist technology company, asked for an hour's worth of audio to recreate her voice.

Sarah and her now adult children hunted for something appropriate – but Sarah lost her voice in 2000 sMart's mobile phones were widely used, and social media captured the moment.

Eventually, an old VHS tape of Aviva as a baby was discovered, filmed on a family video camera in the 1990s. However, the photos were wobbling and the sound was distorted. Those shots tweeted and owned on intense television.

I could hear Sarah's voice eight seconds, barely able to hear.

Sarah was a lively Londoner who worked as a personal assistant in publishing when her life changed.

Married a toddler, the family was hoping for a second baby. But Sarah didn't feel right. Her speech was slowing down, and she felt weak in her left arm.

Unexpectedly, she was diagnosed with MND, sometimes called ALS (amyotrophic lateral sclerosis).

The degenerative condition causes muscle weakness, and of the 1,000 people diagnosed in the UK each year, the NHS says most of them lose the ability to speak.

MND affects men more and shortens life, but it varies from person to person, and some people live longer.

“I was denying it because I thought it was okay,” says Sarah, but “After Eric was born, I got worse quickly.”

Within a few months, Sarah lost her hand use and “all easy to understand speeches.” Her marriage ended soon.

“I was very depressed and I was afraid of disability and death,” she says. Sarah had to resort to 24 hours a day care for her two young children.

“It was hard to see strangers take care of my child, but I'm grateful,” Sarah sat with Aviva and Eric.

Unable to move or communicate easily, Sarah fought isolation. She says she watches bad TV for the first five years and watches her kids grow up.

Eric, 25, says his only memory is that “mum is paralyzed,” but 28, Aviva recalls the moment when he realizes his mother is different.

“I have this memory of asking her to prepare strawberries. She couldn't cut them. She had to ask someone.”

Five years after her diagnosis, the advent of vision techniques ultimately opened up Sarah's communication.

It meant that although she was a synthetic voice like the physicist Stephen Hawking, she could create words and sentences with eye movements.

This technology allowed her to become a volunteer and patronage of charity lifelights for the MND Association.

And she went back to passion, painting, and created her own artwork using her eye-catching technology.

“I was very happy despite my pain and tiredness,” she says.

In Bristol, despite demanding an hour of audio, Smartbox's Simon Poole says his heart sank when he received just eight seconds from a VHS tape.

“I thought you couldn't use that bad audio to create audio,” he says.

But he nevertheless looped it through the latest technology of an international AI-Voice company called ElevenLabs.

The company announced that it would like to offer free audio cloning to one million people who are at risk of losing their speech through situations like MND, cancer, and stroke.

Eventually, Simon was able to use 11 audio isolators to separate Sarah's voice from the sound of the TV that blared. However, the results were thin, with no intonation or personality, and a thin American thing.

So he turned to another app, using thousands of voices to fill in the gaps left by the isolators and to predict where Sarah-like voices accompany intonation.

Eventually, Simon became some audio phrases he was happy with and sent to Sarah.

He recalls the way she told him she was crying to hear her new, old voice first. And one of Sarah's old friends, who knew how she sounded, was “impressed by how realistic it was.”

But how do Aviva and Eric respond when they hear that?

“It was amazing,” Aviva says. “I still agree with that. Now hearing it in everyday life, it still surprises me.”

Sarah's new voice brought her family closer as she was able to express her emotions and communicate them when she was happy, sad or angry. It “makes that difference,” Eric says.

“We can feel who she is as a person,” Aviva says. “My mother is not a disabled person in the corner with a robot that has nothing to do with her.”

Voices created by AI are a major improvement from choices from libraries of old and computerized or recorded voices — says Dr Susan Oman, an AI and social expert at the University of Sheffield.

“It's about your connections between you as an individual and who you are,” she says. “If that's the case [the voice] I don't feel like you, so you don't feel like you. ”

She also says that preserving accents is “really important” at a time when technology can make them homogenous.

“It betrays your class. It betrays your origins. People all over the world are trying to regain lost accents and dialects.”

Sara jokes that she sometimes misses her old composite voice. “I was so gorgeous and people didn't know I was there [really a] Cockney with a slight supple. ”

But she is pleased to have regained her old voice, she says. “I'm happy to be back. It's better than being a robot.”

You can listen to episodes about Sarah about all BBC access via BBC Sound. Subscribe and email your thoughts to accessall@bbc.co.uk