Meta’s AI model replicates voices of loved ones but deems it too risky to release.

Meta, the parent company of Facebook, has unveiled a new generative AI model capable of replicating voices in six languages. Dubbed Voicebox, the technology could benefit visually impaired individuals who can now hear messages from loved ones in their own voices. The model also allows users to speak in foreign languages in their own voice, while creators can edit audio tracks for video or create natural-sounding voices for virtual assistants. Although Meta has not released Voicebox publicly, it noted that the model carries the risks of misuse, and the company aims to find a balance between AI openness and responsibility. According to insiders, Meta has been keen to emphasize its AI-development plans publicly, especially in integrating AI into its products, marking a significant move among Big Tech companies.