Wizard-of-Oz evaluation of speech-driven web browsing interface for people with vision impairments

People with visual impairments typically interact with the Web using screen readers that perform serial text-to-speech narration of the content. Although they rely on keyboard shortcuts to navigate through the content quickly, browsing fatigue caused by too many keyboard presses and clicks, increased cognitive load caused by having to remember many shortcuts, and information overload from having to listen to irrelevant content, are all too common. Speech-based interaction modality has the potential to address these shortcomings by allowing users to engage in a dialog with an intelligent agent capable of translating user commands/requests to system actions and generating appropriate responses to them. This paper presents empirical findings of a Wizard-of-Oz user study conducted with 24 blind subjects to provide a baseline for gauging the usability and effectiveness of speech interfaces for non-visual web access. Specifically, study participants were required to complete a set of typical web browsing tasks using unrestricted speech commands ranging from simple commands such as "click the search button", to complex commands such as "buy this product". Unknown to the participants, these commands were executed by the wizard and appropriate responses were generated with the help of a screen reader. An important byproduct emerging from the study is a new dialog corpus for non-visual web access that will provide pivotal reference data for exploring the design space underlying the development of high performance dialog systems for web accessibility.