Performance Comparison of Conventional and Deep Learning Classifiers for Punjabi Dialect Identification

Dialect identification is the process of identifying the dialect from spoken form and this paper focuses on identifying the four dialects of Punjabi language. For any speech processing activity, the need of database arises at first stage and for Punjabi language, a database primarily for four dialectal variations is constructed. To extract the information from the dataset, Mel Frequency Cepstral Coefficients (MFCC) feature extraction technique is employed along with various pre-processing activities like pre-emphasis, framing and windowing. Then for classification of dialects, conventional classifiers like Support Vector Machine (SVM) and Logistic Regression (LR) are used. Also, Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) are implemented under the category of deep learning classifiers. Performance comparison is done to conclude that deep learning classifiers are giving better outcomes in terms of both accuracy and F1 scores as compared to conventional ones.