Neural Network Fixed size input vectors

When I'm trying to use Neural Networks such as LVQ or Backpropagation , I faced this problem and I spent almost month trying to find good and easy solution.

One method called time normalization using Dynamic Time Warping , it stretching or pressing the input speech duration but it may cause signal distortion.

You can see example written in matlab here : http://www.ee.columbia.edu/~dpwe/resources/matlab/dtw/

There is another good and easy method depend on Window techniques , for more information you can see this thesis Application of a Back-Propagation Neural Network to Isolated-Word Speech Recognition.