The objective of this thesis is to evaluate the feasibility and eventually build a prototype of an automatic system able to support the physicians in the evaluation of mild cognitive decline from speech characteristics, with particular attention to AD. The main features of the system should be: support for Italian language, complete automation from audio transcription to samples classification, low cost and predisposition to be adapted to other languages or pipeline components. Due to the scarcity of data for the Italian language, the machine learning model gives not significant results for this language. However, the same model gives very good results, comparable with the state of the art, on a large English corpus, while being simpler and cheaper than other approaches.