Automated grapheme-to-phoneme conversion system for Romanian

The scope of this paper is to present a novel automated grapheme-to-phoneme conversion system for the Romanian language, based on artificial neural networks together with some transcription results based on a 1.004 words hand built pronunciations dictionary. Using this letter-to-sound transcription system we have also built a large pronunciation dictionary for Romanian language by transcribing the near 140.000 base form entries from the DEXOnline dictionary.