Summary

AI272/speaker is a Codex skill for automating speaker note and presentation script generation from PowerPoint (.pptx) files. It targets academics and researchers, offering slide-grounded narratives by analyzing text, charts, images, and SmartArt, thereby producing comprehensive rehearsal materials and clean speaker notes.

How It Works

This project employs a multi-stage pipeline: structured extraction from PPTX, slide rendering to PNG, OCR, and visual review. These inputs build a "visible element inventory" for each slide, crucial for interpreting complex content like screenshots or SmartArt. This inventory informs the generation of a detailed display version for rehearsal and a clean script, which is injected into the PowerPoint notes pane. This ensures notes are directly tied to visual slide evidence.

Quick Start & Requirements

Installation involves importing the speaker-v7.skill package via a Codex client. Key dependencies include python-pptx, LibreOffice (or macOS Quick Look) for rendering, tesseract for OCR, and python-docx for Word output (Markdown fallback). Vision review capabilities are essential for complex slides.

Highlighted Details

• Comprehensive extraction: Handles text, tables, native charts, and OOXML fallback (e.g., SmartArt).
• Multi-modal analysis: Integrates rendering, OCR, and vision review for a robust visual inventory.
• Dual output: Generates a .docx (or Markdown) display version for rehearsal and injects a clean script into PPTX notes.
• Evidence-based notes: Links spoken content directly to visual slide elements.

Maintenance & Community

The project indicates versioning (speaker-v7.skill), suggesting updates. Specific details on active maintenance, community channels, or notable contributors are not provided.

Licensing & Compatibility

The license type and restrictions for commercial use are not specified. The tool is designed for .pptx files, primarily targeting academic presentation workflows.

Limitations & Caveats

Perfect semantic understanding of all visual elements (images, complex charts) is not guaranteed due to pixel-based data and potential OCR failures. Final notes for visually complex slides depend heavily on the quality of the vision review process. The tool requires explicit confirmation of the output language.