Computational recognition of human emotion using Deep Learning techniques requires learning from large collections of data. However, the complex processes involved in collecting and annotating physiological data lead to datasets with small sample sizes. Models trained on such limited data often do not generalize well to realworld settings. To address the problem of data scarcity, we use an Auxiliary Conditioned Wasserstein Generative Adversarial Network with Gradient Penalty (AC-WGAN-GP) to generate synthetic data. We compare the recognition performance between real and synthetic signals as training data in the task of binary arousal classification. Experiments on GSR and ECG signals show that generative data augmentation significantly improves model performance (avg. 16.5%) for binary arousal classification in a subject-independent setting. CCS CONCEPTS• Computing methodologies → Machine learning; • Humancentered computing → Human computer interaction (HCI).