About This Study

This is an academic research project investigating how well humans can distinguish AI-generated music from real human performances. The results contribute to ongoing research in music perception and AI audio generation.

Privacy & Data

• No cookies, analytics, or third-party tracking of any kind.
• Nickname is used only for the leaderboard. You can use any alias.
• Email is entirely optional — only used if you wish to be contacted for research follow-up.
• IP address is checked solely to prevent duplicate registrations. It is not linked to your responses.
• All gameplay data (answers, scores) is collected anonymously for research purposes only.
• Your data will never be sold, shared with third parties, or used for commercial purposes.
• No AI models run on this site. We only use pre-recorded AI-generated audio clips for the listening test.

Planned Usage

• Academic publication (ISMIR 2026)
• Model validation for AI music detection
• Understanding human perception vs. algorithmic detection

Audio Sources

200 tracks (100 AI + 100 Real), each 12-second segments with variants (8s, 4s, 128kbps, 64kbps).

• AI-generated: Suno (v3/v4), Udio, Stable Audio — collected from public YouTube uploads and direct platform outputs.
• Real (FMA): Free Music Archive — ~25K CC-licensed tracks across diverse genres.
• Real (GTZAN): GTZAN Genre Collection — 1,000 tracks, 10 genres.
• Real (MUSDB18): MUSDB18-HQ — professional multitrack recordings (CC-BY-NC).

All audio is level-normalized (EBU R128), segmented at track level, and filename-hashed to prevent identification bias.

Open Source

This project is fully open source. You can review the entire codebase, including the server, at github.com/unohee/ai-music-survey.

Contact

For questions about this research, please open an issue on GitHub or contact the researcher directly.