Two Level Language Models for Improving the Performance of Tamil Speech Recognition

This paper deals with improving the performance of Tamil speech recognition system, by using two level language models based on phonemes and syllables. The read speech data was normalized and segmented into phoneme sequences using the phoneme segmentation algorithm implemented for Tamil language. A phoneme recognition system converted the segmented phonemes to their corresponding graphemes. The errors in the recognized graphemes were identified and rectified using phoneme based language model. The performance of the recognition system was further improved by combining the graphemes to form the syllable sequences and by applying syllabification rules and syllable based language models to identify the errors in the identified grapheme sequences. .