Cantonese Lion Dance is a highly representative folk performance in the Lingnan region, where drum rhythms play a crucial role in conveying emotions and guiding movements. Traditional performances rely on manual manipulation of the lion head’s expressions, making it difficult to achieve real-time, precise facial expressions that align with the emotional complexity of drum rhythms. With advances in audio emotion recognition within edge computing, new opportunities arise for enhancing the emotional interactivity of lion dance. This study explores how emotion recognition can be integrated into lion dance by automatically analyzing drumbeat emotions and mapping them onto the lion head’s expressions. A real-time CNN-LSTM audio emotion recognition model is trained on labeled drum rhythm data to classify emotions such as “excited,” “tense,” and “calm.” Recognized emotions are mapped to corresponding eye and mouth movements via servos, ensuring synchronized expressions. Implemented on an embedded platform, the system operates with low latency in resource-constrained environments. Results demonstrate that this approach successfully enhances lion dance expressiveness and interactivity, bridging traditional arts with intelligent control. The “drum rhythm–emotion–expression” loop deepens audience engagement, offering new possibilities for cultural innovation and intelligent entertainment, promoting both cultural heritage and social well-being.