Abordagem multidimensional para avaliação da acessibilidade de interfaces vocais considerando a modelagem da incerteza.

Abstract

Voice interaction design per se does not provide quality assurance of the interactive process for visually impaired users. In this dissertation, a method for evaluating voice user interface (VUI) accessibility based upon a set of techniques already well-known to the HCI (Human-Computer Interaction) community is proposed. For each technique, the problem is focused from a different perspective: (i) the user's perspective, which is expressed as views on the product gathered from an inquiry-based approach; (ii) the specialist's perspective, which is expressed by the analysis of the performance results in accessibility testing sessions; and (iii) the accessibility community's perspective, which is expressed by design reviews to determine whether a user interface design conforms to standards. Additionally, Bayesian networks were used in order to make explicit the uncertainty inherent in conformity inspection processes. A case study with DOSVOX system was performed to validate the proposed approach. DOSVOX system was developed at Federal University of Rio de Janeiro (UFRJ) with the aim of helping visually impaired users use the computer. A conformity inspection was performed in accordance with parts 14 (Menu dialogues), 17 (Form-filling dialogues) 171 (Guidance on software accessibility) of ISO 9241. On the other hand, the user performance measurement and the user subjective satisfaction measurement were conducted via accessibility testing. One hundred subjects were enrolled in this study. First, they were categorized as blind (40 users), low vision (20 users) and non-visually impaired (40 users), according to their visual impairment. Second, they were grouped as novices (46 users) and intermediates (54 users), according to their knowledge level in Informatics and experience with the evaluated product. Accessibility test results were statistically analyzed in order to verify the correlation between category performances and between group performances. No statistically significant differences between the user categories or the user groups were found. On the other hand, data comparison showed that the three strategies adopted (user performance measurement, user satisfaction measurement and standard conformity inspection) add to the evaluation process, producing complimentary data that are significant to the process, and reinforcing the relevance of a multi-layered approach for the accessibility evaluation of voice user interfaces.