Publication

Abstract : This study examines the application of machine learning techniques for recognizing and differentiating regional accents in Indian English, with the broader aim of improving the inclusivity and contextual suitability of speech technolo-gies in India's linguistically diverse environment. Using the Indic TTS dataset from IIT Madras, we analyzed speech samples from male and female speakers across eight states, resulting in 16 distinct accent classes. Mel-Frequency Cep-stral Coefficients (MFCC) were employed for acoustic feature extraction, and multiple classifiers-including Support Vector Machines, Logistic Regression, Decision Trees, Random Forest, and XGBoost-were trained and evaluated using stratified 10-fold cross-validation. XGBoost achieved the highest reported accuracy of 99.02 %, followed closely by Random Forest (98.86 %) and CNN-LSTM (98.70 %). These results underscore the potential of accent-aware models to enhance recognition performance and cultural adaptability in speech systems. Beyond improving general speech recognition, the system has applications in accent-specific speech synthesis, personalized voice assistants, call-center automation, and educational tools for language learning and pronunciation feedback.