With the help of artificial intelligence, the voice assistant Alexa can imitate a person’s voice in less than a minute. A professor of computer science warns of the potential for abuse.
In a technical breakthrough by Amazon, a voice could be imitated with extremely little initial data.
(dpa)
Amazon has developed software that can mimic a human’s voice in less than a minute of speech. The online group demonstrated on Thursday how a networked loudspeaker with the language assistant Alexa reads the book “The Wizard of Oz” to a boy with his grandmother’s voice.
The fact that the online company brought up the topic of using the software to preserve the voices of deceased family members caused criticism. Alexa research director Rohit Prasad pointed out that many loved ones had lost in the corona pandemic. “Artificial intelligence can’t make that pain go away, but it can definitely keep the memory alive,” he said before demonstrating the technique. In an interview with the technology blog “Techcrunch”, Prasad then emphasized that the grandmother from the demonstration was definitely alive. “It wasn’t about the ‘dead grandma’.”
Amazon’s technical breakthrough is that the software can mimic a voice with so little initial data. Previously, you had to record text for several hours. The technology is experimental and it’s unclear whether it will reach consumers. A University of Buffalo computer science professor, Siwei Lyu, expressed concern about the software’s potential for abuse. Criminals could pose as family members or someone could mess up stock markets with bogus statements from top managers, he warned at the financial service Bloomberg.