Acoustic Room Impulse Response Simulation with GPUs

Room acoustics simulations have been widely employed to test acoustic and speech signal processing algorithms and, more recently, they have also been used as a data augmentation technique for training machine learning systems which deal with acoustic signals. One of the most common Room Impulse Response (RIR) simulation algorithms is the Image Source Method (ISM), which allows us to get the Impulse Response between two points of a room as the sum of the reflections in its walls; however, the computational complexity of the ISM grows cubically with the length of the reverberation. We present a parallel implementation of the ISM that uses Graphic Processing Units (GPUs) to reduce the simulation time in more than a factor 100 compared with state of the art CPU libraries. We have also included our implementation in a Free and Open Source Python library.