mel-frequency cepstral coefficients (MFCCs) capture: speech features on a perceptual scale

Mel-frequency cepstral coefficients (MFCCs) are derived from a nonlinear "spectrum-of-a-spectrum" representation, which is based on a linear cosine transform of a log power spectrum on a nonlinear mel scale of frequency. This mel scale approximates the human auditory system's response more closely than linearly-spaced frequency bands, allowing for better representation of sound.

The process of deriving MFCCs involves taking the Fourier transform of a windowed excerpt of a signal, mapping the powers of the spectrum onto the mel scale using triangular or cosine overlapping windows, and then taking the logs of the powers. This method captures the perceptual characteristics of sound, making MFCCs particularly useful for applications like audio compression.