HIDDEN MARKOV MODEL BASED SENTENCE CORRECTION SYSTEM FOR URDU LANGUAGE

Abstract

Speech is the most vital mean of communication between humans, as well as humans and machines. Due to resources scarcity not enough work has been conducted for Urdu from the NLP researcher community [2] and a standard application is also needed for immediate response to correct the grammar and pronunciation of Urdu sentence. Hidden Markov Model Based Urdu Sentence Corrector would be designed to correct falsely used every day Urdu Sentences. Speech dataset for Urdu is a fundamental requirement for development on Urdu Automatic Speech Recognition. This research work will be based upon the Urdu data set ofwhich is a medium scale vocabulary of Urdu words. [1] This Final Year Project will address the existing issue ofspeech recognition system in domain ofNLP strategies and machine learning algorithms. We are going to use HMM for our application, which will decrease the time complexity for our application. The previous work was done with two pass parsing, having high time complexity. Our application would also be able to let users hear the correct sentence, rather than just showing it, so they can know the correct pronunciations as well. Also, our application would be efficient enough to correct the any gender specific problem in the spoken sentence.