Improvement in Performance of Tamil Phoneme Recognition using Variable Length and Hybrid Language Models

In this work, the performance of Tamil phoneme recognition system was improved by using language models at the recognition phase. The speech signals were segmented at phonetic level using the language models to identify segmentation points. The phoneme level recognition was done using a similarity measure, based on the acoustic characteristics of the phoneme signal. The errors in the recognized phoneme sequences were then detected and corrected using a language model designed based on the integration of a variable length phoneme language model and inter-word hybrid language model. The performance of phoneme recognition system was improved by using the language models