Design Time Series Time-Delay Neural Networks example errors

I want to utilize the Design Time Series Time-Delay Neural Networks FTDNN example to process doubles instead of cell arrays. Running the example code in the Help section works as expected. However, as a first atempt to use this example to process doubles rather than cell arrays, I used y = cell2mat(y); to get a double array. When I try to execute the example with this mnor modification, I get the error "Inputs and input states have different numbers of samples." when attempting to train the network. Yet, the dimensions of the variable double arrays are clearly the same dimensions as when I run the code with out the cell2mat conversion. I have not seen this error before when training networks using doubles. What could be the source of this?

Variable dimensions when run with y as a cell data type:

>> whos

Name Size Bytes Class Attributes

Pi 1x8 960 cell

ftdnn_net 1x1 43557 network

p 1x592 71040 cell

t 1x592 71040 cell

y 1x600 72000 cell

Variable dimensions when run with y as a double data type:

>> whos

Name Size Bytes Class Attributes

Pi 1x8 64 double

ftdnn_net 1x1 45726 network

p 1x592 4736 double

t 1x592 4736 double

y 1x600 4800 double