Concern about political abuse
OpenAI creates voice cloning software
March 30, 2024, 2:16 p.m
Listen to article
This audio version was artificially generated. More info | Send feedback
Artificial intelligence can write, generate images and videos, speak and soon clone voices. A 15-second recording is enough for this, say the ChatGPT inventors from OpenAI about their latest development. For security reasons, they leave it open when the software will be available to the general public.
ChatGPT developer OpenAI is working on software for cloning human voices – but is holding it back for now due to concerns about misuse. “Voice Engine uses text input and a mere 15-second audio sample to produce natural-sounding speech that closely resembles the original speaker,” writes the US group in a blog entry. OpenAI has been working on the program since 2022, among other things to refine software that converts text into speech.
It remains to be seen whether and when Voice Engine will be available to the general public. Such an application poses considerable risks, especially in an election year, writes OpenAI. Because of possible misuse, the group is therefore approaching the question of approval “carefully and knowledgeably”. OpenAI is hoping for a discussion about the opportunities and risks of the technology and wants to carry out further tests.
In January, fake calls fueled fears of manipulation using artificial intelligence in the race for the White House. In automated calls, a voice that sounded confusingly similar to that of US President Joe Biden called on Democrats in New Hampshire not to take part in the primaries. Such so-called robocalls are a common election campaign tool in the USA.
More reach for videos?
In its announcement, OpenAI also referred to the opportunities of the application. With a partner, OpenAI researched how synthetic voices can support learning to read. Voice Engine also has potential when it comes to translating videos and podcasts, for example, which could then reach a wider audience. In any case, it is important that the cloned voice first reveals itself as being AI-generated.
ntv.de has also been using synthetic voices to read written texts for two years now: In an innovation project, the voice of RTL presenter Maik Meuser, among others, was artificially recreated using self-learning language technology from Microsoft. In order to prevent misuse, a strict “Code of Ethics” is observed when using the votes: The technology may only be used if it is recognizable as such to users of ntv.de. All content that can manipulatively influence opinion formation is excluded. In addition, politicians’ voices are not synthesized.