The feature extraction process is the main step after acquiring speech signals and it is used for developing applications ranging from data compression to complex speech recognition systems. This work aims to describe the development of a system for extracting speech features, based on the Mel frequency cepstral coefficients. The system was designed to be implemented in a portable device, using an embedded system. The paper first describes the general methodology for the calculation of the Mel Frequency Cepstral Coefficients, then it explains each phase individually, and the adaptations performed for the embedded system. In the last part, a comparison of the performance between the embedded algorithm and the one implemented in MATLAB, is presented as a method to validate the results.