AI Recreates Woman's Voice from Eight Seconds of VHS Audio 25 Years After Loss

Artificial intelligence tools, including those from ElevenLabs, have successfully recreated the voice of Sarah Ezekiel, a woman who lost her ability to speak due to motor neuron disease (MND) 25 years ago. The breakthrough was achieved using just eight seconds of scratchy audio from an old VHS tape, as reported by the BBC. This technological feat has allowed Ezekiel to hear and use her own voice again, a significant personal milestone.

Sarah Ezekiel was diagnosed with MND at 34, shortly after becoming a mother for the second time, and lost her speech within months. For years, her communication relied on eye-gaze technology and a synthetic, robotic voice, which her children had grown up hearing. The challenge lay in finding suitable historical audio, as her voice was lost before the widespread use of smartphones and social media.

The "miracle," as Ezekiel described it, began when Bristol-based assistive technology company Smartbox sought an hour of her old audio. Despite the request, only a brief, distorted eight-second clip of her voice could be salvaged from a 1990s family camcorder VHS tape, barely audible amidst background noise. Simon Poole from Smartbox initially doubted the feasibility of voice recreation from such poor-quality and limited input.

However, by leveraging ElevenLabs' advanced AI technology, specifically their Voice Isolator, Poole managed to separate Ezekiel's voice from the surrounding sounds. Although the initial result lacked intonation and personality, further processing with other AI applications, trained on thousands of voices, helped to fill in the gaps and predict natural vocal nuances. The final reconstructed voice was then presented to Ezekiel.

Upon hearing her own voice again, Sarah Ezekiel expressed profound emotion, stating, "When I first heard it again, I felt like crying. It's a kind of miracle." Her children, Aviva and Eric, also reacted with amazement, noting that the recreated voice allowed them to "feel who she is as a person." This personalized voice offers a significant improvement over generic synthetic voices, emphasizing the importance of individual identity and connection in communication.