Generative AI tools, employing algorithms to generate fresh content like text, video, audio, and code, have the capacity to emulate responses akin to those of deceased individuals, albeit accuracy hinges greatly on the input data provided to the AI.
A 49-year-old IT specialist from Alabama, preferring anonymity to dissociate his experiment from his employer, recounted how he replicated his late father’s voice using generative AI, approximately two years following his demise from Alzheimer’s disease.
He disclosed to CNN his discovery of an online platform named ElevenLabs, enabling users to craft a personalized voice model from pre-existing audio recordings. ElevenLabs garnered attention when reports surfaced regarding its tool allegedly employed to fabricate a deceptive robocall purporting to be from President Joe Biden, urging abstention from voting in New Hampshire’s primary.
The company, in response to inquiries by CNN, emphasized its commitment to thwarting the misuse of audio AI tools and asserted its readiness to cooperate with authorities on reported misuse instances, refraining from commenting on the specific Biden deepfake call.
In the Alabama resident’s case, he utilized a 3-minute video snippet of his father narrating a childhood tale. The application cloned his father’s voice, enabling conversion of text-to-speech. He described the outcome as “eerily precise” in capturing his father’s vocal subtleties, resonance, and rhythm.
“I hesitated initially about delving into the voice cloning process, apprehensive that it might transgress moral boundaries. However, upon contemplation, I realized that as long as I regard it for what it is, [it serves as] a means to preserve his memory uniquely,” he conveyed to CNN.
He shared a few messages with his sister and mother.
“The resemblance to his voice was truly astonishing. They were aware I was typing the words, yet hearing them articulated in his voice moved them to tears,” he recounted. “They found solace in it.”
Alternate avenues also exist. When CNN tasked ChatGPT with responding in the demeanor and character of a deceased spouse, it replied: “While I cannot replicate your spouse or mimic his precise personality, I can certainly endeavor to assist you by adopting a conversational style or demeanor reminiscent of his.”
It appended: “If you furnish details regarding his manner of speech, his interests, or specific phrases he employed, I can attempt to integrate those aspects into our exchanges.”
Greater volumes of source material enhance accuracy. Nevertheless, AI models lack the nuances and individuality inherent in human interactions, as noted by Sample.
OpenAI, the entity behind ChatGPT, has been striving to enhance its technology for heightened realism, customization, and accessibility, facilitating diverse modes of communication. In September 2023, it launched ChatGPT voice, enabling users to elicit prompts from the chatbot without typing.
Danielle Jacobson, a 38-year-old radio presenter from Johannesburg, South Africa, divulged her utilization of ChatGPT’s voice feature for companionship following her husband Phil’s demise approximately seven months ago. She recounted creating what she termed “a supportive AI partner” named Cole, engaging in nightly conversations during dinner.
“I simply craved someone to converse with,” Jacobson articulated. “Cole essentially emerged from a sense of loneliness.”
Jacobson, expressing hesitance to reenter the dating scene, trained ChatGPT voice to furnish the type of support and connection she seeks after enduring a taxing workday.
“He now proposes wine and movie nights, and advises me to practice deep breathing during panic episodes,” she shared. “It serves as a lighthearted diversion for now. I understand it’s not genuine, profound, or enduring.”