What kind of input values from MFCC should be used to train Neural Network for speech recognition?

I have set of .wav files and using MFCC i have extracted 13 MFCC coefficients for each sound file(.wav). For a particular .wav file i got an m-by-n matrix where n is the 13 MFCC coefficients and m varies for different .wav files. EG: file1.wav=>54-by-13 matrix file2.wav=>65-by-13 matrix

For training Neural network i should have a fixed m-by-n matrix size.Further i calculated correlation coefficient for each m-by-n MFCC matrix[eg: corrcoef(54-by-13 matrix)] and obtained 13-by-13 matrix for all the .wav files uniqely.Is this the right way to apply correlation cofficient for varying MFCC coefficient matrix inorder to obtain a unique m-by-n matrix for all the .wav files?