The result of applying a speech analysis methodology, often as the input stage to a speech recognition system, that is designed to reflect the way we hear. The speech input is sampled, then windowed (see windowing), and a fast Fourier transform is taken from which the power spectrum is calculated. This is arranged into mel-spaced frequency bands (see mel) and the logarithm of the amplitude of each band is taken. A discrete cosine transform is then taken to produce the MFCCs. Although this output is referred to as a cepstrum, it should be noted that the two are not directly equivalent.