Increasing Agency of Screen-Reader Users in Consuming Information From Online Data Visualizations

This paper addresses a fundamental equity problem in data visualization accessibility: screen-reader users are limited to consuming only the information that visualization creators choose to provide, with no agency to customize what they hear. Prior research shows screen-reader users extract information 61% less accurately and spend 211% more interaction time than sighted users with visualizations. The authors present VoxEx, a system that enables screen-reader users to specify their preferences for how they consume information from online data visualizations across three modalities — alt-text (summary content and verbosity), data tables (sorting preferences), and sonification (audio graphs with pitch type, sound type, and speed). VoxEx comprises three components: a configuration portal (built with React and Apollo) where users log in via Google authentication and set preferences, a Chrome browser extension that stores preferences in session storage and makes them available to web pages, and a backend server (Node.js with GraphQL and Postgres) for secure preference storage. The system was informed by two formative studies: a survey of 60 screen-reader users and semi-structured interviews with 12 users. Seventy-five percent of survey respondents wanted to customize alt-text, 73% wanted data table customization, and 48% wanted sonification customization. VoxEx integrates with VoxLens, an existing open-source multi-modal JavaScript plug-in for visualization accessibility, enabling creators to implement user preference support through simple conditional if-else logic in their code.

In a five-day diary study with three screen-reader users, participants recognized improved efficiency in information extraction, with one noting it allowed them to "do things quickly and efficiently, get the information, glance at it like a sighted person." Another emphasized the importance of autonomy: "Being able to choose the information felt really important and meaningful because I had more autonomy and control over the information I wanted." NASA-TLX scores showed low workload: low mental demand (M=3.0), low physical demand (M=1.3), low temporal demand (M=5.0), low effort (M=5.3), low frustration (M=3.7), and high perceived performance (M=17.3). No participants reported privacy or security concerns. In the creator case study with three visualization creators, all rated implementation difficulty as minimal ("1" on a 1-7 scale), with one stating "it doesn't seem like that much work." All three found VoxEx useful for all users, not just screen-reader users — including people with color vision deficiency. Areas for improvement included adding definitions for configuration options (subsequently implemented as accessible tooltips) and domain-of-interest-based verbosity levels (implemented using ChatGPT to categorize web pages). Participants highlighted VoxEx's potential as an educational tool for explaining statistical concepts like standard deviation.

This research makes a significant contribution to accessibility by reframing the problem from "making visualizations accessible" to "giving users agency over how they consume information." The current paradigm where creators decide what information screen-reader users receive through a single alt-text description creates an inherently unequal experience. VoxEx demonstrates that user agency can be implemented with minimal creator effort through a preference-sharing architecture that uses browser session storage as a bridge between user preferences and creator code. For accessibility practitioners, the key design pattern is powerful: a centralized preference portal, a browser extension that persists preferences across sites, and simple conditional logic for creators to respect those preferences. The finding that customization benefits extend beyond screen-reader users to people with color vision deficiency and learning disabilities suggests that user agency in information consumption is a universal accessibility principle. The import-preferences feature, which allows users to import their existing screen reader settings, reduces configuration burden and demonstrates thoughtful attention to the real-world workflows of assistive technology users.