Currently, the computational methods in vibroacoustics are mainly based on discretization like the finite element method (FEM). The main drawback of these conventional methods is that the computational effort considerably increases with finer discretization. Especially applications become computationally expensive, where a high number of model evaluations is required, e.g. uncertainty quantification, optimization or inverse problems. In contrast, an ever-increasing amount of data is being collected and is available, which strengthens the need for efficient data-driven surrogate models. Deep learning (DL) is an emerging machine learning method, but a purely supervised approach requires a large amount of high-quality training data, which are often not available.In this contribution, physics-informed neural networks (PINNs) are used in order to solve vibroacoustics problems in a data-driven manner. The idea is to incorporate the residual of the partial differential equations that govern the elastodynamics behavior into the loss function of the neural network, which guides the training of the network to only learn physically admissible solutions. This additional physics-informed prior information allows for better generalization even in the small data regime. The application of PINNs is demonstrated on academic elastodynamics models and its performance is discussed in comparison to conventional neural networks.