Kernel convolution model for decoding sounds from time-varying neural responses