The present paper explores the use of several deep neural network architectures to carry out a grapheme-to-phoneme (G2P) conversion, aiming to find a universal and language-independent approach to the ...