Brain-computer interfaces are artificial output channels that allow direct communication between the brain and a computer or machine without the use of the muscular system. In the health field has a high potential to improve the quality of life of people with motor disabilities. To address this challenge, in this research we present a novel method based on power spectral density (PSD) with functional data to the recognition of silent speech syllables with electroencephalography signals. The method use brain signals of language area with the specific task of thinking the respective syllable. The classification process for five silent speech syllables (/fa/, /pa/, /ma/, /la/, /ra/), in Spanish, was carried out with a support vector machine, achieving an accuracy between 67.78% and 72.67%.