Assessing Closed Captioning Quality Using a Multilayer Perceptron

Abstract

Closed Captioning (CC) is a telecommunication service to provide Deaf or Hard of Hearing (D/HOH) audiences the text equivalent of what hearing audiences experience in TV. The quality of CC is often interpreted as an accuracy and assessed in the empirical measure of counting number of errors. Although the regulators necessitate certain rules in the factors of CC quality, the D/HOH community members, who are the primary audiences to the CC, are not completely satisfied. One solution to solve this issue can be including the perspective of D/HOH audience in the assessment process. This research made an attempt to design an automatic quality assessment system for CC using artificial neural networks-multilayer perceptron trained from the D/HOH audience's subjective ratings, and the representative values extracted from the caption file, and the transcript file. As an initial stage of research, the trained model was then compared with other statistical regression models.