Approaches to Address AI-enabled Voice Cloning
Approaches to Address AI-enabled Voice Cloning Today, the FTC announced four winners of the Voice Cloning Challenge, which was launched to address the present and emerging harms of artificial intelligence, or “AI”-enabled voice cloning technologies.
Approaches to Address AI-enabled Voice Cloning
- Today, the FTC announced four winners of the Voice Cloning Challenge, which was launched to address the present and emerging harms of artificial intelligence, or “AI”-enabled voice cloning technologies.
- The FTC received submissions from a wide range of individuals, teams, and organizations.
Leveraging solutions to provide upstream prevention or authentication
- Prevention or authentication refers to techniques that limit the application and misuse of voice cloning software by unauthorized users.
- One commonly discussed approach for prevention and authentication is watermarking, which often refers to a “broad array of techniques” for embedding an identifying mark into a piece of media to track its origin to help prevent the misuse of cloned audio clips.
- [2] Invisible or visible watermarks can be altered or removed, potentially rendering them unhelpful for differentiating between real and synthetic content.
Applying solutions to detect solutions in real-time
Real-time detection or monitoring includes methods to detect cloned voices or the use of voice cloning technology at the time during which a specific event occurs. Studies reveal a spectrum of efficacy for voice cloning detection solutions.[5] The effectiveness of such solutions is especially important when considering the types of AI-enabled voice cloning scams – such as fraudulent extortion scams – that the technology can enable.
Using solutions to evaluate existing content
- The post-use evaluation of existing content includes methods to check if already-created audio clips such as voice mail messages and audio direct messages contain cloned voices.
- One potential way to evaluate existing audio clips is to develop algorithms that detect inconsistencies in voice cloned clips.
Looking forward: Preventing and deterring AI-enabled voice cloning scams and fraud
- While there are many exciting ideas with great potential, there’s still no silver bullet to prevent the harms posed by voice cloning.
- Further, voice service providers – telephone and VoIP companies – need to continue making progress against illegal calls.
- In addition, the Commission has recently enacted a new Impersonation Rule, which will give the agency additional tools to deter and halt deceptive voice cloning practices.
- There is no AI exemption from the laws on the books and the FTC remains committed to protecting consumers from the misuse of “AI”-enabled voice cloning technologies.
[2]https://arxiv.org/pdf/2306.01953.pdf
[3]https://www.technologyreview.com/2023/08/09/1077516/watermarking-ai-trus...
[4]https://www.banking.senate.gov/imo/media/doc/voice_cloning_financial_sca...
[5]https://arxiv.org/pdf/2307.07683.pdf; https://arxiv.org/pdf/2005.13770.pdf
[6]https://arxiv.org/pdf/2308.14970.pdf
[7]https://arxiv.org/pdf/2402.18085v1.pdf
[8] https://arxiv.org/pdf/2307.07683.pdf; https://arxiv.org/pdf/2402.18085v1.pdf