Musical Metamerism with Time--Frequency Scattering
This AI can generate music that sounds the same but is fundamentally different.
Deep Dive
Researchers have developed a new AI method called "musical metamerism" that can generate audio fragments that sound identical to human ears despite having completely different underlying waveforms. The technique uses joint time-frequency scattering via the open-source Kymatio Python library, requires no manual preprocessing like transcription or beat tracking, and works directly on any audio recording. This creates perceptual illusions where two different musical signals are indistinguishable to listeners.
Why It Matters
This could revolutionize audio watermarking, music copyright protection, and our understanding of human auditory perception.