Keyword spotting requires a small memory footprint to run on
mobile devices. However, previous works still use several hun-
dred thousand parameters to achieve good performance. To ad-
dress this issue, we propose a time delay neural network with
shared weight self-attention for small-footprint keyword spot-
ting. By sharing weights, the parameters of self-attention are
reduced but without performance reduction. The publicly avail-
able Google Speech Commands dataset is used to evaluate the
models. The number of parameters (12K) of our model is 1/20
of state-of-the-art ResNet model (239K). The proposed model
achieves an error rate of 4.19% , which is comparable to the
ResNet model.