Dear Ser/Madame,
A basic approach would be to plot the frequecy spectrum with the function fft of your voice sample and use a mean square error (in MATLAB you could use for example the command 'mean(errors.^2)') to compare this voice sample to multiple saved samples, which are already recorded and saved. Also when you have found your coefficients, you then could compare those to the coefficients of the sample you want to compare with.
For more information how an 'Isolated Word Recognition System in MATLAB' could be developed, can be found on this Mathworks website here.
Good luck! Christiaan
