Average decoding accuracies of the acoustic parameters in layers S5, S6, and S7 with different kernel sizes.

Note that S5 with kernel size 10×10 and S6 with kernel size 10×10 are layers in the original network. Layer S5 with kernel size 20×20 was obtained by fixing layers S1 to C4 of the original network; layer S6 with size 5×5 and layer S7 with kernel size 5×5 were obtained by fixing layers S1 to C5 of the original network. The STRF sizes in these layers are indicated in parentheses.