Nvidia has clarified what data it’s using to train its generative AI Ace microservice, which saw an expansion at CES this year. The company now confirms that its tools are “commercially safe” and only trained on data it has the rights to.
Nvidia’s announcement was met with skepticism earlier this week when the company gave Digital Trends an ambiguous answer about data usage. When we asked what the program was trained on (specifically citing what it uses to generate voices), Nvidia said that there was “no simple answer” and that it’s hard to pin down due to the number of different tools Ace uses. The comments sparked speculation on social media, with many speculating that the data included copywritten material. Nvidia now confirms that’s not the case.
“NVIDIA ACE microservices, Audio-2-Face and Riva Automatic Speech Recognition, are commercially safe, trained on data licensed from trusted partners as well as commercially licensable open-source datasets,” Nvidia tells Digital Trends in an email clarifying its original statement. “Similar to most work in the voice acting and motion capture industry, NVIDIA pays fully up front for the ongoing use of voice and mocap data.”
Digital Trends followed up to confirm in clear terms that Ace is only trained on data that Nvidia owns or has the rights to, with no scrapping of copywritten material outside that pool. The company confirmed that.
We’ve also reached out to Convai, a third-party partner behind some of Ace’s AI tech, to confirm that. We’ll update this story when we get a response.
