Say It All: Feedback for Improving Non-Visual Presentation Accessibility

Document Title

Say It All: Feedback for Improving Non-Visual Presentation Accessibility

Document Information

• Subject Area: Accessibility Technology and Presentation Design
• Keywords: Accessible presentations, real-time feedback, blind accessibility, image description, speech-visual alignment, user study

Research Background and Problem Statement

• Problems and Challenges:

  • In slide-supported presentations, visual content is often not fully described, which affects the understanding and engagement of blind and visually impaired audiences.
  • The authors analyzed 90 presentation videos (containing 610 visual elements) and found that 72% of the visual content was not adequately described.
  • Current accessibility presentation guidelines lack specific tool support, and the authors highlight significant challenges in practical implementation, such as translating abstract guidelines into actionable practices.

• Research Significance:

  • Visual presentations are a critical medium for conveying information in educational, research, and business settings. However, if the content is not accessible to all audiences, this communication is incomplete.
  • Enhancing the accessibility of slide presentations enables blind, visually impaired, and other constrained audiences to receive information equally.

• Research Motivation and Related Work:

  • Inspired by prior feedback tools and systems for accessible media creation (e.g., accessibility tools like Lift and WAVE for web design), the authors aim to adapt these methods for real-time and post-presentation improvements in slide content.
  • The study explores algorithms for aligning auditory and visual content and the learning effects in interdisciplinary scenarios, incorporating existing accessibility presentation guidelines (e.g., VG1-7 simplified rules for audio-visual alignment).

Solution

• Methods and Solutions:

  • The authors propose "Presentation A11y," which provides real-time and post-presentation feedback through multiple interfaces:
    1. Real-Time Feedback Interface:
       • Transcribes the speaker's narration in real time and analyzes whether the visual elements on the slides are covered.
       • Highlights slide elements that have not been described.
    2. Post-Presentation Feedback Interface:
       • Offers page-by-page content summaries and specific suggestions, such as removing unmentioned text or simplifying overly complex media content.
  • The system uses Optical Character Recognition (OCR) and label generation algorithms to classify and tag text and images on the slides.

• Innovations:

  • Embeds high-level accessibility guidelines into specific tools, providing element-level real-time feedback and detailed post-presentation suggestions.
  • Supports precise alignment of speech and visual content, addressing the issue of insufficient text and media descriptions.

• Implementation Steps and Technology:

  1. Perform DOM analysis on slides to extract visual elements.
  2. Use Google Cloud's real-time speech transcription and image labeling API to parse presentation content.
  3. Generate improvement suggestions and textual element description scores for different presentation stages (real-time and post-presentation).

Research Outcomes

• Key Results:

  • In experiments, participants using Presentation A11y significantly improved their coverage of slide content.
  • Compared to traditional interfaces, text coverage increased by 11%, and media coverage improved by 25%.
  • With post-presentation feedback, participants identified 3.26 times more accessibility improvement points.

• Advantages:

  • Assists speakers in providing more comprehensive descriptions of visual elements, enhancing the accessibility of presentation content.
  • Improves user focus and target identification for improvements through element-level feedback.

• Experiments and Evaluation:

  • A user study involving 16 participants tested the system in various contexts (classrooms, conferences, lectures).
  • Quantitative data showed that using Presentation A11y significantly improved visually impaired users' self-assessed presentation accessibility (average score increased from 4.00 to 5.06).
  • Qualitative feedback indicated that participants felt the system raised their awareness of accessibility and helped optimize their narration content.

• Limitations and Future Directions:

  • Limitations:
    • The current feedback interface primarily targets sighted users and does not fully serve blind and visually impaired presenters.
    • Quantifying the quality of descriptions for complex media and highly contextual content remains a technical challenge.
  • Future Directions:
    • Develop accessibility content creation tools that are friendly to blind authors.
    • Extend the system to hybrid presentations in physical environments (e.g., using gestures, interacting with the audience).
    • Enhance image analysis technologies to strengthen automated feedback for complex media (e.g., applying DenseCap to generate richer content descriptions).