This report explores the potential of grapheme as a modeling unit for acoustic modeling in Amharic, Tamil and Telugu. While the three languages are considered phonetic, Amharic has an exception where in syllables of the form /CVC/ may be orthographically represented as /CC/. Here, the context determines the identity of the vowel within the syllable. We employ a transcription correction model to handle this exception to see how much information the Grapheme unit is inherently capable of capturing. For the two south Indian languages, which are relatively more exception free, the comparison is done against the phonetic baseline system. Graphemes are a promising unit for acoustic modeling as borne out by our preliminary results.