ChatGPT creator says Voice Engine can replicate an individual’s voice based mostly on a 15-second audio pattern.
OpenAI has unveiled a instrument for cloning individuals’s voices however is holding again on its public launch as a result of issues about doable misuse in a key election 12 months.
Voice Engine can replicate an individual’s voice based mostly on a 15-second audio pattern, in accordance with an OpenAI weblog put up demonstrating the instrument.
However the ChatGPT creator is “taking a cautious and knowledgeable method” to the know-how and hopes to begin a dialogue on “the accountable deployment of artificial voices”, the corporate mentioned within the weblog put up printed on Friday.
“We acknowledge that producing speech that resembles individuals’s voices has critical dangers, that are particularly high of thoughts in an election 12 months,” the San Francisco-based start-up mentioned.
“We’re participating with U.S. and worldwide companions from throughout authorities, media, leisure, training, civil society and past to make sure we’re incorporating their suggestions as we construct.”
We’re sharing our learnings from a small-scale preview of Voice Engine, a mannequin which makes use of textual content enter and a single 15-second audio pattern to generate natural-sounding speech that carefully resembles the unique speaker. https://t.co/yLsfGaVtrZ
— OpenAI (@OpenAI) March 29, 2024
OpenAI mentioned it might make “a extra knowledgeable determination” about deploying the know-how at scale based mostly on testing and public debate.
The corporate added that it believes the know-how ought to solely be rolled out with measures that make sure the “authentic speaker is knowingly including their voice to the service” and stop the “creation of voices which might be too much like distinguished figures.”
The misuse of AI has emerged as a serious concern forward of elections this 12 months in nations representing about half the world’s inhabitants.
Voters in additional than 80 nations, together with Mexico, South Africa and the USA are going to the polls in 2024, which has been dubbed the largest election 12 months in historical past.
The affect of AI on voters has already come beneath scrutiny in a number of elections.
Pakistan’s jailed former Prime Minister Imran Khan used AI-generated speeches to attraction to supporters within the run-up to the nation’s parliamentary elections in February.
In January, a political operative for the long-shot US presidential candidate Dean Phillips put out a robocall impersonating US President Joe Biden that urged voters to not forged their ballots in New Hampshire’s Democratic Celebration major.
OpenAI mentioned it had applied a number of security measures for its companions testing Voice Engine, “together with watermarking to hint the origin of any audio generated by Voice Engine, in addition to proactive monitoring of the way it’s getting used”.