We present DeepBeam, a hybrid model that combines traditional beamformers with a custom lightweight neural network. The former reduces the computational burden of the latter and also improves its generalizability, while the latter is designed to further reduce the memory and computational overhead to enable real-time and low-latency operations. Our   Read More ...