In recent years, music recommendation systems have evolved to incorporate user-centric factors for more personalized experiences. This study proposes an innovative approach that leverages Speech Emotion Recognition (SER) to dynamically recommend music based on the emotional state inferred from users’ spoken input. By integrating a 2D CNN-LSTM hybrid model trained on emotionally rich audio datasets, the system accurately detects emotional cues such as happiness, sadness, anger, or calmness. Hyperparameter tuning and advanced regularization techniques were applied to improve performance and generalization. The final model achieved a test accuracy of 93%, demonstrating strong capability in classifying emotional states from speech and maintaining robustness under varying acoustic conditions. These recognized emotions are then mapped to suitable music genres or tracks, enhancing the user’s emotional alignment with the recommended content. The integration of SER into music recommendation systems presents a novel path toward context-aware and emotionally adaptive multimedia services.