Perceiving AI in Music: Human Evaluation of AI-Generated Melodies and AI Detection Sensitivity

Oehler M, Oldach J, Zwißler F. Perceiving AI in Music: Human Evaluation of AI-Generated Melodies and AI Detection Sensitivity. In: 2025 AES International Conference on Artificial Intelligence and Machine Learning for Audio. Audio Engineering Society; 2025.pp. 27:1-9. Available from: https://aes.org/publications/elibrary-page/?id=23016

@inproceedings{Oehler2025_23016,
  author = {Oehler, Michael and Oldach, Jasper and Zwißler, Florian},
  title = {{Perceiving AI in Music: Human Evaluation of AI-Generated Melodies and AI Detection Sensitivity}},
  booktitle = {2025 AES International Conference on Artificial Intelligence and Machine Learning for Audio},
  pages = {27:1-9},
  year = {2025},
  month = sep,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=23016}
}

TY  - CPAPER
TI  - Perceiving AI in Music: Human Evaluation of AI-Generated Melodies and AI Detection Sensitivity
AU  - Oehler, Michael
AU  - Oldach, Jasper
AU  - Zwißler, Florian
T2  - 2025 AES International Conference on Artificial Intelligence and Machine Learning for Audio
SP  - 1
EP  - 9
M1  - Paper 27
PY  - 2025
DA  - 2025/09/02
UR  - https://aes.org/publications/elibrary-page/?id=23016
PB  - Audio Engineering Society
LA  - en
AB  - Recent advancements in generative artificial intelligence (AI) have led to increasingly sophisticated music composition systems, yet their perceived creative quality and detectability remain debated. This study examines human perception of AI-generated music across two paradigms: a melody continuation task, where AI- and human generated melodies were rated on four aesthetic dimensions, and an AI detection task, where participants distinguished between AI-composed and human-composed songs. Results indicate that AI-generated melodies were consistently rated lower than human compositions, with no significant differences between AI models or the impact of expert feedback. In the detection task, participants identified AI-generated songs at a level above chance, though sensitivity remained low. Notably, musical sophistication, as measured by the Gold-MSI, did not significantly influence ratings or detection accuracy. These findings highlight persistent limitations in AI-generated music while raising questions about the perceptual boundaries between AI and human creativity.
ER  -