Development of a Language Model to Enhance the Performance of Hindi Automatic Speech Recognition

Automatic Speech Recognition (ASR) performance depends on numerous issues. Automatic Speech Recognition (ASR) is a self-contained, automated procedure for deciphering and transcribing spoken language using machines. An ASR system commonly accepts audio input from a speaker via a microphone, evaluates it using a pattern, model, or algorithm, and generates an output, typically in the form of text. The language model facilitates the transformation of these probabilities into coherent linguistic expressions. The language model utilises statistical analysis of training data to assign probability to words and phrases. A language model is created by computing the probabilities of n-grams. The Automatic Speech Recognition (ASR) system is applied to both the pre-existing language model and the newly created language model in order to generatea transcriptedoutput. This output is then utilised to determine the Word Error Rate. A language model is essential in contemporary natural language processing(NLP) applications. In this article, the implementation of the ASR using N-gram model and the improvement in the word error rate by considering an language model is discussed.