Meta Launches ‘Voicebox’: A Text-to-Speech AI That Learns Similar to ChatGPT

Meta AI has recently launched Voicebox, a new text-to-speech generator that has the ability to learn in-context. The company claims that the resulting audio can be up to 20 times faster than other comparable AI models. Unlike traditional text-to-speech architecture, Voicebox employs an architecture similar to OpenAI’s ChatGPT or Google’s Bard, and can generalise through in-context learning. This means that it is possible for Voicebox to translate text to speech, remove unwanted noise by synthesising replacement speech, and even apply a speaker’s voice to different language outputs. Meta has developed a tool for detecting if speech was generated by its technology, and has made plans to mitigate possible use of fake audio in the future.