n this paper we present our argument that context information could be used in early stages i.e., during the definition of mapping of the words into sequence of graphemes. We show that the early tagged contextual graphemes play a significant role in improving the performance of grapheme based speech synthesis and speech recognition systems.
