Meta, formerly known as Facebook, has introduced a groundbreaking AI tool named ‘Voicebox’ that showcases significant advancements in AI-powered speech generation. However, the company has decided against releasing it to the public due to the potential disastrous consequences it could entail.
According to a blog post by Meta, Voicebox has the ability to generate audio clips of speech in six European languages, surpassing other competing AI models in various aspects. What sets Voicebox apart is its capability to perform tasks beyond its original training, a feature that Meta claims makes it highly superior to its counterparts.
The tool’s primary function is to accurately replicate a person’s voice through text-to-speech conversion using as little as a two-second audio sample. While this may seem innocuous, it possesses significant destructive potential in the wrong hands. Voicebox could potentially be utilized for malicious purposes, including the creation of fake revenge pornography or even triggering international conflicts.
Public figures, especially politicians, have numerous audio recordings available online, making it relatively easy to compile speech clips and utilize Voicebox to produce eerily realistic imitations of their voices. While similar tools already exist, their level of credibility falls short. Social media has witnessed amusing videos featuring fabricated scenarios of political figures like Joe Biden, Donald Trump, and Barack Obama playing video games together. Although these videos offer entertainment, the audio quality is far from convincing and would not deceive anyone with common sense.
Nevertheless, Meta believes that Voicebox possesses the potential to deceive a significant majority of people. Consequently, rather than releasing the tool to the public, Meta has chosen to publish a research paper and provide a classifier tool that can identify speech generated by Voicebox, differentiating it from real human speech. Meta describes the classifier as “highly effective,” although not infallible.
While acknowledging the potential for misuse and unintended harm associated with Voicebox and similar tools, Meta highlights the potential benefits that AI speech generation can bring in the future. Voicebox has the potential to provide more naturalistic speech to individuals who are mute or have difficulty communicating, removing barriers to interaction. Additionally, it could facilitate real-time translation, bringing us closer to the concept of “universal translator” devices depicted in science fiction.
Furthermore, Voicebox offers other practical applications. Meta explains in its blog post that the tool can be used to edit and enhance recorded speech. For instance, if a word was mispronounced or background noise interrupted the audio, Voicebox can isolate the problematic segment and seamlessly ‘re-record’ it using the original speaker’s voice. This feature showcases impressive capabilities while also evoking a sense of slight unease.
Meta’s cautious and deliberate approach to Voicebox is commendable. Microsoft’s haste in integrating Bing AI into various platforms has resulted in controversy, and OpenAI’s release of ChatGPT has generated numerous peculiarities over the past year. As we find ourselves in the midst of an AI gold rush, with these tools permeating all aspects of our lives, it is essential to exercise responsible and thoughtful deployment.
