By Ishvaraus Davis

The music industry is representative of culture in many ways. Look no further than artists like Beyoncé generating over $4,000,000,000 on a single tour. Recently, one of the most popular artists in the world, Drake, released a song entitled Taylor Made. The song interestingly begins with a verse from Tupac, followed by a verse from Snoop Dogg. Both of these verses are, in fact, AI clones. More specifically, they are voice conversions that are powered by artificial intelligence. The technology is in the spotlight now due to the cultural influence of music, but the technology will have ramifications well-beyond songs.

What is voice conversion?

Voice conversion (or singing voice conversion) has a very straightforward objective: transform the source voice (voice A) into a target voice (voice B). In that way, anyone can sound like Denzel Washington, Barack Obama, or others with clips of audio available online. This technology has been around for years, but it’s never been easier to make realistic voice clones than it is now. Here’s an example:

How to Make Your Voice Sound Like Drake, The Weeknd, Taylor Swift and More using AI

Tools such as uberduck.ai streamline the capability to transfer one voice to another voice. In addition to easy to use products, there are enough papers and Github resources available online to make this technology easily accessible to anyone that’s willing to take the time to find it. With that being the case, we are quickly entering a world where digital audio can’t be trusted. Think of the amount of times a day that we use someone’s voice as a mechanism for trust. We participate in phone calls, Zoom calls with the video off, voice messages, audio-only social media, dating apps, music apps, and more. There are a number of interactions that open people up to being tricked, manipulated, and taken advantage of, especially as large swaths of the population are unfamiliar with this technology. Given the current state of the digital literacy divide, Black communities are at risk in particular.

We have to look no further than this 82-year old that sent $17,000 to scammers that imitated his son-in-law’s voice. Or the Joe Biden robocall that used voice cloning to dissuade voters from turning out at the New Hampshire elections. Or the numerous other scams that lost people $1.1 billion in 2023 alone. The majority of these scams were not using sophisticated AI conversion tools so that number is likely to increase substantially in the coming years. The potential risk is exacerbated even more when deepfakes are added to the equation. Just this week, Microsoft put out a paper on creating deepfakes from just 1 image and an audio clip. It’s no doubt that we will enter a world very soon in which the generated and the real are indistinguishable. So, how do we protect ourselves and those around us?

At an individual level it all begins with education. Once people are aware of these capabilities, and how they might be used, they are much more resilient to the potential people that want to take advantage of them. Beyond that, we need the tools that people can use to identify what’s legitimate and what isn’t. The Federal Trade Commision took a step in that direction with their announcement of the recent winners of their Voice Cloning Challenge. The winners proposed using AI to determine if audio is AI generated, adding small changes to voice audio that would make it difficult for AI to replicate, and even using sensors at the time of recording to determine if audio is coming from a human. This is in tandem with their recent rulemaking that prohibits individuals from posing as businesses or government agencies. This rulemaking, however, does not go far enough. Tennessee recently signed into law the Ensuring Likeness Voice and Image Security Act. This law protects against the unauthorized use of an artist’s voice from AI. Everyone deserves these protections, whether we’re in the artistic space or not.