Musimo: An Intelligent Music Analysis and Emotion Recognition System

Musimo, an intelligent music analysis and emotion recognition system designed for sound engineers, video editors, and music composers. This paper presents MASE-Net, a novel MultiScale Attention Squeeze-Excitation Network designed for continuous valence and arousal regression in music audio. The proposed architecture integrates parallel multi-scale convolutional blocks with squeeze- excitation channel attention, bidirectional long short- term memory networks, and multi-head self-attention mechanisms to capture hierarchical temporal and spectral features. The system also enables similarity search, personalized recommendation, and future integration of user libraries. Experimental evaluation shows that hybrid CNN- Transformer models outperform baseline architectures in emotion classification accuracy and feature embedding quality.