Abstract: Biological neurons exhibit nonlinear and often chaotic dynamics that enhance adaptability and efficient information processing. Building on these concepts, this study explores how chaotic dynamics can be integrated into convolutional neural networks (CNNs) to enhance speech emotion recognition. Using the RAVDESS dataset–both in its original and augmented versions–the research compares three different architectures: a baseline CNN employing the traditional ReLU activation, a CNN using a fixed logistic chaotic activation, and another CNN incorporating a trainable chaotic parameter. Data augmentation improved model generalization, while chaotic activations introduced additional flexibility, with the trainable variant achieving the highest overall balance across emotional classes. The findings reveal that chaotic activations mitigate class-level imbalance and reproduce adaptive response behaviors similar to biological neurons. This work does not aim to achieve state-of-the-art accuracy, but to demonstrate how controlled chaos can serve as a biologically inspired mechanism for enhancing neural computation. By linking nonlinear dynamics observed in brain activity with artificial learning systems, the study provides a conceptual bridge between neuroscience and artificial intelligence in the context of emotion recognition.