This is the repository for a Pytorch-based implementation of FRA-RIR method for data augmentation purpurse in simulating reverberant signals. FRA-RIR is a Image-source method (ISM) based RIR simulation method that replaces the explicit calculation of the delay and reflections of the virtual sound sources by random approximations. With a CPU-only simulation pipeline, FRA-RIR can be significantly faster than other RIR simulation tools, enabling fully on-the-fly data simulation during training, but also improves the model performance when evaluated on realistic RIR data.
Please refer to our paper on Arxiv for details.
