Taking into account the implicit phonotactic constraints of the Italian, a graph-oriented computational approach to the grapheme-to-phoneme transcription of written texts is described. The system is based on the mathematical theory of Finite States Automata, generalized and augmented to obtain a simple syntax-directed translation schema. The rules are totally based on the pure orthographic form of the words, and no level of grammatical knowledge of their classes is considered. A particular set of exceptions is often associated to the whole set of rules and, owing to the modular quality of the automaton, an optimized rule-exceptions check mechanism is built up, albeit in a preliminary and incomplete form.