ldenoue

I've adopted a very simple approach: 80 words per "paragraph". I am now experimenting with computing the embeddings of each sentence and try to detect topic segments. But the simple approach yields pleasant segments AFAIK.

The pdf should show them as bold. I’ll see how to add the highlights to the markdown and html. Hard for text (maybe use * around each?)

You can share the link as it contains the list of timestamps corresponding to your highlights.

So others or yourself can see them again.

The first time the deep net needs to load. I should add a progress of some sort. Thanks

Thanks happy you like it. Send us feedback for new ideas

You can do that with ScreenRun (which I developed) https://screenrun.app/

I've tried using LLMs to restore and clean up raw unpunctuated transcripts, but they tend to hallucinate new words. And chunking is an issue since long transcripts need to be split according to the LLM input context size. But where do we split if we don't have the punctuation yet?

In Scribe I chose instead to restore punctuation marks using a token-classifier (here a DistilBert running in your browser) https://www.appblit.com/scribe

Laurent

You might enjoy my app (iOS only) to load and save the video and show you the transcript so you can only focus on that video https://www.appblit.com/scribe

How’s traction with Pop? Love the landing page with the animated cursor that gives a preview of what the product does!

Still using it when updating my apps written pre swift but 100% Swift now (not SwiftUI though) https://www.appblit.com/

I’m also doing this but it adds punctuations, paragraphs and chapter headers because most raw YouTube transcripts lack proper punctuation

https://www.appblit.com/scribe

Perhaps you’ll like Screegle then because it overlays your webcam on what you share as ONE stream and you can share one or several windows without revealing your entire desktop https://www.appblit.com/screegle

The tweet links to the zip file to install from source https://www.appblit.com/static/scribe-extension.zip

Free YouTube on your phone here with transcripts, audio and videos in landscape, all stored for offline listening, reading and watching https://www.appblit.com/scribe

Back when my kids were 9 we made a simple math game called PopMath in html and it was using a basic animation loop to move bubbles around the screen. I then converted it into native ObjectiveC for the iPhone and it was a hit with them!

Nothing fancy: https://apps.apple.com/us/app/popmath-lite/id303258911

It’s using web technologies so yes it could also be on Android. I’ll see what can be done.

I wrote an app called PDF Reflow that reflows the original PDF using image processing to cut out words into tiles so you see the reflowed version of the text in their original look.

https://www.appblit.com/pdfreflow

this is an example video with Ilya but you can use the front page to paste any Youtube link or search Youtube.

The raw transcript is punctuated in your browser using a small DistilBert token-classifier using the JS version of HuggingFace transformers library using Xenova

I also wrote a web app but it adds missing punctuations and chapters to further enhance readability. The AI models run locally in your browser, no ChatGPT used at all.

An iOS app is also coming soon so you’ll be able to listen and read while offline

https://www.appblit.com/scribe

If you want to try scribe I developed last week, it adds punctuations from the raw YouTube transcripts so you can read them more easily. It also adds chapters every 3 paragraphs to more easily skim the content. All runs in your browser using 2 models https://www.appblit.com/scribe

It uses the existing YouTube transcript and adds punctuations using a deep net that runs in your browser

Chapters should come first then?

Yes I’m also surprised how many tasks can be done using ONNX in the browser. The model I found and converted for browser use is here https://huggingface.co/ldenoue/distilbert-base-re-punctuate

It’s not open source but I did publish the ONNX converted model to HuggingFace here https://huggingface.co/ldenoue/distilbert-base-re-punctuate

It’s amazing what Xenova Transformers JS can do to run ONNX models. And WebGPU is coming soon!

I would like first to generate chapters (perhaps using a T5 model). Summarization is also possible.

Most YouTube transcripts lack punctuations. So this app adds them back using a deep learning model based on DistilBert converted to ONNX for inference in your browser.

Is the app also handling PDF documents for the reader mode? I made an app related to this for iOS called ReaderView where you can also highlight your read for later pages, including PDF documents. (No images are preserved for pdf files yet, only the text)

I noticed a lot of YouTube transcripts do not contain punctuations, which makes them hard to read. Fortunately, HuggingFace has several DistilBert token classifiers trained specifically on adding punctuations, so I converted a small enough model to ONNX so it can run in our browser (Using the fantastic Xenova Transformers JS library).

I would like to also add summaries and chapters using ONNX. Any idea welcome!

You can use QuickTime on MacOS to see your iPhone as a video. Not sure other platforms.

This app is free. In addition to recording your windows with sound, you can hide the cursor. The app was developed as a companion app to ScreenRun https://screenrun.app/ to help you record high quality videos of any window.

Web browsers do not have the capability to get where you click when using getDisplayMedia so this native app saves the clicks in the description field of the MP4.

So when you paste or import that video into ScreenRun website, the web app reads the metadata and is able to automatically create for you the points where zooming effects should happen.

Even if you don’t use the ScreenRun website you can use this native macOS app to just record screencasts

You can also check ScreenRun to create beautiful videos from screenshots or recordings in seconds https://screenrun.app/