Artificial Intelligence-based enhancements supporting Linguistic E-Learning System

Abstract

The Automated Phonetic Transcription - grading tool (APTgt) is a web-based interactive linguistics E-learning system developed by the Auburn HCI group for teachers and students in the Department of Communication Disorders (CMDS) at Auburn University. This system aims to provide interactive phonetic transcription exams to help students recognize and translate disordered speech or speech without disorders. The teacher functions are to create online exams for students and the system automatically generates grades with the grading tool. In my research, based on the APTgt system, I built an auto exam generator that can significantly reduce the time for teachers to create the exams. Also, a word bank with a classification function was developed to classify the input, phonetic words, into six difficulty levels and store them in the database for future use. To build a more intelligent system and enrich the performance of the APTgt system, I also developed machine-learning/deep-learning based functions to the system to address our research questions as follows: • The Classification and Regression Tree based classification module in work bank. • The MFCC and DTW / convolutional neural network-based speech disorder classification as an additional feature for the system. • The Transformer based Grahame-to-Phoneme(G2P) converter which can help teachers better create phonetic words exams from normal English words. • The encoder-decoder based IPA-to-Speech speech synthesis model to provide the pronunciation of the words/sentences in IPA format.