Development of an Automatic Speech Recognition System for Irula Language: A Machine Learning Approach

The Irula language, which is predominantly spoken by the Irula tribes in southern India, offers specific challenges because of its linguistic features and limited resources. To address these challenges, this research study presents the development of an Automatic Speech Recognition (ASR) system for Irula, which accelerates the conversion of spoken language into English text. In the proposed system, the data is collected and pre-processed, features are extracted, and acoustic and language models are developed for efficient translation. The spectral subtraction noise removal method is used to reduce the background noise from collected speech data during the data pre-processing. The Mel Frequency Cepstral Coefficients (MFCC) are then used to extract the features which represent the characteristics of speech signals. Acoustic modelling is implemented using Hidden Markov Models (HMMs) and their fine-tuned version. A pre-trained language model such as GPT is used for accurate translation. Finally, the web application is developed for translating Irula language into English text to increase the accessibility for Irula speakers.