This dataset contains high-fidelity "soft labels" for 6,870 music tracks, specifically curated for training lightweight Music Emotion Recognition (MER) models. The labels were generated using LAION-CLAP (laion/clap-htsat-fused) acting as a Teacher model.

🏗 Project Architecture

The goal of this project is to distill the knowledge of a heavy, billion-parameter transformer (CLAP) into a smaller "Student" model that can run… See the full description on the dataset page: https://huggingface.co/datasets/beastLucifer/sangeetkar-labelled-data.