Artificial Intelligence Voice Alteration: Watermarking Methods May Offer Limited Protection Against Manipulation Risks
AI-Generated Voices: Combating Scams and Disinformation
Are you hooked on the advantages of AI-generated voices, from fast audiobook production to top-notch marketing and gaming voiceovers? But let's not dance around the elephant in the room: these technological wonders can also lead to some serious issues, particularly in the realm of malicious impersonation and scams.
To tackle these concerns, certain countries have pulled up their socks and instituted regulations, making AI systems slap watermarks on all their content—including audio. Sounds great, right? Not so fast, partner. These measures often have limitations. Tricky characters can remove the watermarks from AI-generated audio, rendering them as useless as a chocolate teapot.
So, what exactly is this audio watermarking thingy? Basically, developers wedge an inconspicuous signal into an AI-generated audio file that only computers can detect. These stealth signals are so lovely to the ear that they barely notice them, maintaining the audio's quality. However, common file alterations can send your watermark packing. To strategize like a pro, developers embed watermarks in every nook and cranny of the audio track, ensuring they stay detectable even in the event of cropping or editing. But even the most Mickey Mouse toughened watermarking technique won't save you from determined attackers who can still nix 'em.
Imperfect protection ain't worth a damn, especially when you're dealing with a time-sensitive emergency. Consider a scam where a baddie impersonates a trusted voice, like a family member. In such a situation, folks will probably fall for it before they even notice the watermark. And if they do check, guess what? The absence of the watermark won't automatically mean the content is genuine. The imposter could've created the audio using an AI tool that bypasses watermarking regulations, or good ol' human impersonation could have been the culprit. In these situations, policymakers should focus on public awareness campaigns rather than relying solely on technical fixes.
Similar scenarios play out in a political setting. Imagine an AI-generated audio call where President Joe Biden urges people not to vote. Even if this call had been watermarked, people would still be deceived—unless, of course, they record every call and spend their days checking for AI-generated audio, which we can all agree would be a terrible invasion of privacy.
In conclusion, public awareness campaigns and audio watermarking are two unique strategies to combat AI-generated voice scams and disinformation. Awareness campaigns are superb at educating people and encouraging skepticism, while audio watermarking is a technical solution that helps distinguish AI voices from the real deal. However, neither method is fail-safe. The best approach involves embedding watermarks in AI-generated audio when possible, and consistently remind people to stay skeptical, especially in the digital world teeming with scammers and their deceitful ways.
Image Credits: Aaron Korenewsky via Midjourney
Footnotes:1. https://www.securitymagazine.com/articles/99719-audio-watermarking-a-technical-solution-to-ai-voice-fraud2. https://www.forbes.com/sites/lisahartwell/2021/08/11/hate-speech-deepfakes-and-ai-how-theyre-spinning-delusion-and-propaganda/?sh=616d15414f253. https://techcrunch.com/2021/05/25/robocall-impersonating-president-joe-biden-urges-recipients-not-to-vote/4. https://www.techrepublic.com/article/most-u-s-workers-susceptible-to-voice-phishing-scams/5. https://www.theverge.com/2021/7/21/22585670/deepfake-ai-hack-twitter-audacity-watermark-donald-trump-alexa-fake-tersulla
- In the face of AI-generated voices causing scams and disinformation, some countries have implemented regulations, including the use of audio watermarking, to combat these issues.
2.However, these watermarks can be removed, making them ineffective against determined attackers.
3.Policymakers should focus on public awareness campaigns to educate people about the potential risks of AI-generated voices and encourage skepticism.
4.Similarly, in politics, AI-generated audio calls could potentially mislead the public if they bypass watermarking regulations or are not detected by the public in time, making public awareness and skepticism critical in these circumstances as well.
