This paper presents the impact of data augmentation on the performance of biometrics gait system. The carried out work demonstrates the concept of using generative models in the form of data distribution modelling neural networks as data augmentation mechanism. The work proposed an artificial neural network architecture capable of modeling the normal distribution of Inertial Measurement Unit (IMU) readings. The proposed models were capable of simultaneous synthetic generation of triaxial accelerometer and triaxial gyroscope signals. Effectiveness of the proposed augmentation mechanism was compared with performance on the Riemannian Hamiltonian VAE (RHVAE) and timeVAE models. Validation of the experiments was carried out using the author's corpus of human gait cycles (100 subjects) collected over two days. The study resulted in an increase in the metric of the biometrics system from a baseline of 0.73 ± 0.019 to 0.775 ± 0.014 F1-score. Proposed neural network architecture allowed to outperform the competing models (RHVAE), for which an F1-score metric of 0.753 ± 0.016 was achieved.