The lecture presents various aspects of automatic speech processing (SP), from spoken contents extraction to high level categorization of speech signals. We show how machine learning provides solutions to the main issues that the speech processing systems have to deal with. Speech is one of the main way that humans communicate together. It is a complex process that involves, in a highly integrated way, perception abilities and cognitive processes. In spite of the efforts produced by the scientific community for simulating these abilities, knowledge-based approaches failed in modeling of speech. Nowadays, most of the SP methods relies on statistical modeling of speech. The lecture presents this theorical framework in which the major issues in speech processing are formulated. Then, the main tasks of SP are overviewed, especially speaker identification and speech recognition and understanding.