VATA: An Improved Personal Computer-Based Vowel Articulation Training Aid
A. Matthew Zimmer · ODU Digital Commons (Old Dominion University) · 2024
This thesis discusses work accomplished on the further development of a vowel articulation training aid for persons with hearing impairments. The system produces visual feedback about the quality of articulation for ten American English monophthong vowel phonemes. A large database of vowel recordings has been collected and used to improve and test the recognition rates of artificial neural networks used by the training aid. A maximum-likelihood classifier has been introduced into the system to improve performance by reducing the likelihood of incorrect feedback produced by the displays. A series of experiments has been performed to examine effects of tuning parameters, and of varying the amount and type of speech data used to train the system. Much of the program code for the speech display system has been revised to use a more modular structure, which in turn provides for easier maintenance and testing. Current system features and operational details are discussed in an appendix.