This project used CNN for music emotion recognition with spectrograms. This repository contains everything needed including dataset generation, experiment execution and simple visualization. The experiments are based on DEAM dataset, result shows some CNN models can have top level performance compared to LSTM models, even for continuous annotation prediction.
If you think it's helpful, you can get started reading the report and examples. The code is written simply for experiment purposes with limited time. It's quite handy to use for me, but don't expect it to be too expandable and reliable. The license is only for code. For the report, all rights reserved.