speaker
Generate grounded academic speaker notes from PowerPoint decks using vision and OCR.
Combine text extraction, slide rendering, OCR, and vision review to write grounded PPTX notes.
Speaker is a Codex skill that reads a PowerPoint deck, extracts structure and text, renders slides to PNG, and runs OCR and vision reviews. It produces timed rehearsal notes and injects them directly into the PowerPoint notes pane for academic presentations.
What it is
The tool builds a visible element inventory covering text boxes, native charts, tables, screenshots, and embedded images. It budgets speech length using a pause-aware pacing model for both English and Chinese, then generates display documents and injects clean notes back into your `.pptx` file.
Who it's for
It suits academic researchers, professors, and professionals who present complex slide decks and need precisely timed, evidence-grounded notes. It is not built for casual users who just want generic text summaries without checking visual slide elements.
Key features
Pause-aware pacing model
Budgets speech length with a deterministic model that accounts for slide transitions and pause marks.
Region-scoped OCR
Runs optical character recognition only on picture and media regions with an automatic full-slide fallback.
Compact vision review
Compiles a shared review packet to evaluate visually complex slide content with a vision-capable agent.
Notes injection
Writes clean speaker scripts straight into the PowerPoint notes pane and exports a complete rehearsal document.
Up and running in 3 steps
Our honest verdict
What's great
- Captures charts, legends, tables, and text embedded inside images that traditional text extractors miss.
- Enforces strict per-slide word budgets to prevent your presentation from running over time.
- Generates both a rehearsal document and a notes-injected PowerPoint file automatically.
- Provides an OOXML fallback to extract SmartArt and grouped-shape text not exposed by standard libraries.
Where it falls short
- Requires a Codex client or Claude Code environment to run as a skill project.
- Slide rendering depends on external tools like LibreOffice or macOS Quick Look.
- Involves managing multiple intermediate JSON and work files generated during extraction.
Pricing & license
The tool is free to use under its repository setup, though the exact license is not stated in the repository metadata. The only cost is running it yourself within your local environment.
Open remote AI-training roles
767 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
What tools does slide rendering rely on?
It attempts to use LibreOffice or soffice first, falling back to macOS Quick Look when available.
How does it handle presentation timing?
It applies a deterministic pacing model budgeted at roughly 110 wpm for English and 165 characters per minute for Chinese.
Where are intermediate evidence files stored?
They are stored inside a local work directory alongside your output deliverables.
