Those Audio Summaries, Are they AI Generated?

Yes.

Mike's Audio Summaries, is a YouTube podcast experiment.

Google Gemini created image: GoogleLM on a laptop

I've published three at time of writing this and they have caused much amusement, confusion, and conversation.

So how does each episode get produced, here's the complete run down of how I do it.

1: Write a blog post

It all starts with me having an idea of blog post that will be lengthy, something that will take me a while to write as it is fact heavy, has many sections, and needs me to actually do some research. 

The posts I use for the AI audio summaries are definitely not based in my usual posts, they just don't have enough in them.

2: Use NotebookLM

Once I've written my post and it is published I then go to Google's NotebookLM: https://notebooklm.google/

NotebookLM (Google NotebookLM) is a research and note-taking online tool developed by Google Labs that uses artificial intelligence (AI), specifically Google Gemini, to assist users in interacting with their documents. It can generate summaries, explanations, and answers based on content uploaded by users. It also includes "Audio Overviews", which summarizes documents in a conversational, podcast-like format.

[source: Wikipedia]

2.1 "Upload" content into a new notebook

Once logged in I then create a new notebook, ("folder"), and upload all the content I want NotebookLM to base its workings on.

For me so far that has been almost exclusively web page links and it definitely starts with my own blog post, obvs. I then add into the notebook all the webpages I have referenced.

Currently NotebookLM has a limit of 50 sources per notebook but even with my heaviest blog posts I haven't got close to it.

It's interesting that a small number of web pages I upload are not used because the publishers have disallowed it, Stuff for instance.

2.2 Ask a few 'sanity' questions

As the first page, my original blog post, is referenced by NotebookLM is generates a whole series of things in what's called a "Notebook guide":

  • A text summary
  • Suggested questions
  • Notes, which are answers to questions asked

I type in a few questions just to check that it can answer the basics.

So far so good, in essence it's an AI chatbot but based only upon the content I have given to it. If you ask it about anything else not present on the webpages referenced then it won't have a clue.

2.3 Create and download "Audio overview"

This is the gold, the bit that everyone is talking about.

I push the "Create audio overview", wait about 5 minutes as the world's water, power, and other resources are sucked up and used, before downloading the generated .WAV file created.

I could just listen to it in NotebookLM if course but I've got bigger ideas.

3. Format and Covert To MP4

A .WAV file is just audio, one of the simplest and oldest media formats there is tbf. I hunted around for a system to convert to .MP4 so that I could load it up into YouTube to share with the world and embed back on my original blog post. 

I current use Veed as it not only converts but allows me to add background images (often created with Google's Gemini, but not always), text, and most fun an actual sound wave.

4. Upload to "Mikes Audio Summaries" YouTube podcast

The conversion is done,  download as an .MP4 and uploaded into my new YouTube podcast playlist, Mikes Audio Summaries.

5. Share the YouTube video

I also then re-edit the original blog post and embed the YouTube video as well as being chuffed to bits and blasting out the link to mates etc.



So, there you have it, how I generate an AI audio summary video on my selected blog posts. 

How long I'll do this for is up for grabs as it is an experiment, it'll be irregular at best.

I'm now off to create and add one to this very post ...

Other posts with audio overviews:

Comments