Google's AI Lyria model to embed inaudible watermarks in music
What's the story
Google's DeepMind plans to incorporate undetectable watermarks, known as SynthID, into music produced by its AI Lyria model.
This includes tunes generated with YouTube's latest audio creation features.
These inaudible watermarks will enable users to recognize AI-generated music without affecting the overall listening experience.
According to a DeepMind blog post, the watermark should remain imperceptible to human ears and still be identifiable even after the audio track undergoes compression, speed changes, or additional noise.
Details
SynthID as a safeguard against generative AI harms
Watermarking technologies like SynthID are deemed essential in addressing potential risks linked to generative AI.
US President Joe Biden's executive order on artificial intelligence emphasizes the need for new government-led standards for watermarking AI-created content.
Although this is a promising field, existing technologies have yet to offer a foolproof solution against counterfeit content.
What Next?
DeepMind's unique approach to audio watermarking
DeepMind describes SynthID's audio application as a process that transforms the audio wave into a 2D visual representation, illustrating how the frequency spectrum of a sound changes over time.
This groundbreaking method is said to be "unlike anything that exists today," distinguishing it from other watermarking technologies.
As AI-generated content becomes increasingly common, tools like SynthID will be crucial in maintaining transparency and authenticity in the digital realm.