In a detailed and critical analysis of the theoretical foundations of automatic speech recognition (ASE) and of their basis, the fundamentals of signal phonetics, it is shown that an unnecessary prerequisite is regarded as a constitutive condition. That is why the aim of ASE appears as a decoding of phonetic information. Contrarywise, it is attempted here to recognize the problem of ASE as the automation of a process of active phonetic information retrieval through iterative signal processing. For this purpose the construction of a complex vector structure is suggested which takes into account the phonological and linguistic knowledge of the relevant means of oral communication as well as the similarities and regularities that can be found in the signal domain. The process of active information retrieval is then defined by a system of rules which, with the help of this structure, relates incoming (communicative) signals to a text and which at the same time constantly adapts the vector structure on the basis of such relations. In view of this active retrieval procedure the ASE phoneme or word identification by static class features turns out to be a spurious and erroneous simplification.