Dataset · mippia
Benchmark for AI music generation tracking
mippia/HAIM
An audio and text benchmark in English and Korean for evaluating neural network roles in music production.
What's inside
The dataset contains 1 to 10 thousand examples with audio and text in Korean and English. Unlike simple human-versus-AI splitting, data is labeled by specific track production roles, including composer, lyricist, and other stages. Created as a benchmark to track generative AI like Suno and Udio in the music industry.
What you can do with it
- Evaluating generated music detectors by individual production stages
- Training audio classification models to identify AI composers and lyricists
- Comparing synthetic audio analysis algorithms
- Tasks
- Audio classification
- Languages
- Korean, English
- Size
- 1K–10K rows
- Downloads this month
- 76.9K
- Likes
- 0
How to load it
from datasets import load_dataset
ds = load_dataset("mippia/HAIM")