Meta, the parent company of social media platforms Facebook and Instagram, said that it has created the world’s most powerful artificial intelligence-powered voice generator.
However, the company said it would not be releasing the technology to the public yet because it’s simply too dangerous.
The technology, called Voicebox, is a breakthrough in generative AI technology for speech, according to Meta, as it’s gone beyond what it was “specifically trained to accomplish.”
Other popular generative AI platforms such as Bard and ChatGPT are able to generate text based on various prompts. Voicebox can do the same thing, but it can also create audio completely from scratch, or it can modify a sample in various different styles.
Meta published a blog post recently that said that Voicebox is able to create audio clips in six different languages, and is able to outperform all the other generative AI tools on the market.
That being said, the company announced it would not be releasing Voicebox to the public – either through its code or a model – just yet because of “potential risks of misuse.”
The company wrote in its blog post:
“While we believe it is important to be open with the AI community and to share our research to advance the state of the art in AI, it’s also necessary to strike the right balance between openness with responsibility.”
All that Meta has released at this point in regard to Voicebox is a few audio samples as well as a research paper that details the approach it takes and the various results it’s been able to achieve.
The company added that it’s been able to develop a “high effective” feature that distinguishes between audio that’s generated from Voicebox and authentic human speech.
The technology has gone through more than 50,000 hours of training on recorded transcripts and speech from audiobooks that are publicly available in the languages of English, Portuguese, Polish, German, Spanish and French.
It seems that Meta is working hard to create new generative AI technology that makes it very hard to distinguish between real human content and that generated by technology – which would seem to fly in the face of what it says it’s trying to do in terms of content moderation on its two social media platforms.
Last week, the company said that it has a new model of AI that creates images that are “indistinguishable” from ones that are created by humans.
As many other AI technology companies are putting a halt on the development of their programs because of the inherent dangers they present, Meta continues to push forward.
Just last month, major technology companies such as Google, Microsoft, DeepMind and OpenAI all signed a statement that equated the risks that AI technology provides to that of wars and pandemics. The executives at Meta, though, dismissed those warnings and wouldn’t sign onto the statement.
Instead, Meta is creating features for generative AI that could be eventually usd in consumer products, such as a tool that would be able to modify photos posted by users on Instagram using text prompts.
