In order for systems to react intelligently to user emotions, emotion detection is a crucial component of human–computer interaction. This study introduces a deep multimodal framework that integrates speech and gesture models to enhance emotion recognition. The proposed framework utilizes deep long short-term memory (LSTM) networks to process speech and gesture inputs, achieving superior performance compared to existing benchmarks. The framework achieves accuracy of 91.4% on a benchmark dataset, by employing a multimodal feature fusion technique to improve emotion recognition accuracy. Children’s emotion dataset was used for to ensure the model effeteness for demographic, encompassing both basic and complex emotions. Due to developmental diversity and complexity of children’s emotions, a greater challenge is to recognize emotions of children’s compared to adults. Compare the proposed model with existing model which demonstrates the model reliability and efficiency. This study focuses on emotion detection system for younger population, offering applications in many areas like child development, healthcare, social inclusion, safety and monitoring, education, and emotional well-being.