Music Emotion Recognition (MER) and generation are rapidly evolving areas in affective computing, enabling systems to understand and produce music aligned with human emotions. However, existing methods often struggle with capturing long-term temporal dependencies and context-specific emotional nuances, especially when processing complex audio and lyrical content. Additionally, they cannot often generalize effectively across diverse musical genres and emotional spectrums. To address these limitations, we propose EMOGEN : Emotion-aware Music recognition and Generation using Enhanced Neural networks . EMOGEN employs a bidirectional Long Short-Term Memory (Bi-LSTM) network combined with an attention mechanism to extract and emphasize key emotional features from musical audio signals. The framework also incorporates emotion conditioning for generative models, enabling the creation of emotionally coherent music. The proposed system is designed for use in emotion-aware music recommendation platforms and therapeutic music generation, adapting music outputs based on user moods or targeted emotional outcomes. Experimental results demonstrate that EMOGEN significantly improves emotion recognition accuracy and generates music that is more consistent with target emotions than baseline models. This highlights its potential to enhance user experience and emotional well-being through adaptive, intelligent music systems. EMOGEN achieves 94.9% accuracy, 97.6% emotional coherence, 90% genre adaptability, 88.1% cultural sensitivity, a lowest 7.2% error rate, and the fastest 4.0s processing time, demonstrating exceptional robustness and efficiency.
Tao Zhang (2026) studied this question.