Enhanced semantic communication schemes for speech signals

Abstract

Two new models for semantic communication systems are proposed. The first model incorporates the convolutional block attention module, which considers attention techniques in both the channel and spatial domains. The second model applies the efficient channel attention (ECA) network with reduced complexity. Experimental results demonstrate that the convolutional block attention module‐equipped model improved signal‐to‐distortion ratio performance by at a signal‐to‐noise ratio of while maintaining a similar number of parameters compared to the existing model using squeeze‐and‐excitation network. Meanwhile, the efficient channel attention‐equipped model reduced parameters by approximately without any degradation in performance compared to the existing model.