Within the scope of the Man-Machine Communication PRC (Programme de Recherches Concertees) Speech topic, we have developed the GEPH System, which is meant to enhance BDLEX, the lexical database of both spoken and written French. GEPH is supposed to derive the permissible pronunciations of a given statement, from its underlying representation, as this is secured from both analysing the statement for syntax and consulting BDLEX for a phonological representation of words. Within the scope of the ESPRIT SAM Project (Speech Assesment Methodologies), the semi-automation of the annotation tasks, applied to speech corpora, involves as a sub-task an automatic phonotypical transcription. In the context of an expert system in phonology, this makes the problem easier for only one typical pronunciation is supposed to be derived. Furthermore, this pronunciation is to remain abstract and will not require applying low-level rules. In our paper, we intend to give the configuration of the automatic transcription system worked out from GEPH and made to operate on compatible IBM-PC -the system is written in C language.