We improve on a piece-wise linear model of the trajectories of Mel Frequency Cepstral Coefficients, which are commonly used as features in Automatic Speech Recognition. For this purpose, we have created a very clean ...