Reproduce “Echoes of Humanity: Exploring the Perceived Humanness of AI Music” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Flavio Figueiredo, Giovanni Martinelli, Henrique Sousa, Pedro Rodrigues, Frederico Pedrosa, Lucas N. Ferreira
A study from UFMG explored how listeners discern AI-generated music from human compositions using an "in-the-wild" dataset and a randomized controlled crossover trial. Researchers found that while general distinction is challenging, listeners' ability to correctly identify AI music rises to 66% when AI and human songs are highly similar, relying on vocal, sound, and technical cues.
Key insights
- Listeners' ability to reliably distinguish AI-generated music from human-made music is significantly enhanced when songs are perceptually similar in style.
- Practical musical experience and prior familiarity with AI music generation services are factors that positively correlate with correct identification.
- Qualitative analysis revealed that listeners primarily rely on vocal, sound quality, and technical production cues, along with lyrical coherence, to discern AI-generated tracks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.