AlQasmi القاسمي
العربية Contact me
Work
Speech to text

Recording transcription and summaries

Turns a recording into a transcript linked to the slides, then into an organized summary.

An example I built: Study

Nothing said goes unwritten: every sentence is linked to the slide it was said on, no camera needed.

Study interface: today's page with the latest recording and where to resume

The idea

Lectures carry a lot that the slides and the book don't: explanations, clarifications, examples. I built Study so none of it gets lost.

I also wanted to understand every slide: what it says, and how the lecturer explained it. So I linked the speech to the slide it was said on, without a camera.

How it works

  1. 1

    Transcribes the recording with accurate timestamps, in a first, direct pass.

  2. 2

    In a second pass, the model reviews the text and fixes wrong words from context, using the course notes and their terms.

  3. 3

    Links every stretch of speech to the slide it was said on, so you read the slide and its explanation together.

  4. 4

    Summarizes, extracts the key points, adds the important quotes from the book, and draws the key diagrams inside the summary.

  5. 5

    All of it runs locally on my machines.

Whisper large-v3-turbo Cohere Transcribe Arabic PaddleOCR-VL Qwen 3.8 27B Runs locally
Each recording
References
Recording
Speech as said
First pass
Timed transcript
Second pass
Fixes from context
Course notes, book and slides

Terms for fixes, slides for linking, the book for quotes in the summary
Summary linked to the slides
Key points, book quotes and diagrams

Let's turn your organization's recordings into knowledge you can look up

Contact me
Next projectTools I built for myself