Text-to-Pronunciation conversion is often used for speech synthesis and speech recognition-related systems. In this paper we present a data-driven, language-independent and multi-stage model for Text-to-Pronunciation conversion. With a Grapheme/Phoneme pair well aligned dictionary for training and utilizing a re-scoring strategy for those graphemes likely to be tagged erroneously, our model can not only increase the efficiency but also achieve a high accuracy than other data-driven approaches that have been applied to the same tasks. Keywords: Text-To-Pronunciation, Grapheme, Phoneme.