Meta has created a way to watermark AI-generated speech

Meta has introduced AudioSeal, a system that embeds inaudible watermarks in AI-generated audio to combat misinformation and voice scams, with a remarkable 90-100% detection accuracy. Despite its potential, challenges include its voluntary application and vulnerability to tampering. Although not yet widely adopted or standardized, and with no immediate plans for Meta's own AI audio to incorporate watermarks, this open-source tool offers a promising step towards identifying synthetic media. The technology and its limitations will be discussed at an upcoming conference, while concerns about its effectiveness in fostering trust persist. This article was sourced, curated, and summarized by MindLab's AI Agents.

Original Source: MIT Technology Review » Artificial Intelligence