Yes, we know what that headline sounds like, but unfortunately it's true. The company Meta, which owns Facebook, Instagram or for example WhatsApp, said it has created a new artificial intelligence system that can generate speech in six different languages at a level that is indistinguishable from real human speech. The speech can be generated in different styles, inflections and even emotions. However, Meta also adds that it will not bring this artificial intelligence to the market for fear of possible risks.
Meta's speech generator is called Voicebox and can be set to produce different outputs in different styles. Currently, the tool can speak six languages and supports a number of functions such as noise removal. Meta claims that Voicebox is much further apart from previous speech generating systems, mainly thanks to the Flow Matching system, which allows the system to learn the speech of normal people and thus train on diverse data.
However, Meta also states that the risks that this system may carry are so high that it prefers not to list it to avoid abuse. A Meta spokesperson did not specify specific abuse possibilities, but commented on the matter as follows: “As with other powerful AI, we recognize that this technology also brings potential for abuse.”
ProThe problem is quite clear, however, because AI can generate the voice of specific people without their consent and can thus help in creating fake videos, for example in news reporting. Thanks to similar systems, it is not a problem to record a video in which, for example, the president of any country declares war on another country and no one will be able to tell the difference between whether this situation really happened or whether the video is created by AI.
“There are many exciting uses for generative speech models, but due to potential risks, we will not be releasing the Voicebox model or code at this time. However, we believe it is important to be open and share challenges in the AI community to advance the field of AI. However, it is important to find the right balance between openness and accountability,” Meta representatives said. You can listen to Voicebox demos right here.
This is good news especially for banks that have implemented voice authentication. It seemed stupid to me when I think Česká spořitelna came up with it. You say hello and they recognize you by your voice. I doubt the system can handle such a voice generator.