I’ve recently been following a few Facebook Groups related to genealogy and AI (Artificial Intelligence) tools. I’m a complete AI novice but someone in the group mentioned NotebookLM so I thought I’d take a look.
NotebookLM is an AI-powered research and note-taking tool by Google Labs. You upload your own sources such as research notes, website addresses, etc. to the web interface and it helps you summarise, analyse, and interact with your own uploaded documents and content. Google say your personal data is never used to train NotebookLM.
I’m somewhat impulsive and haven’t read all of the documentation on how it works but I noticed a really cool feature called Audio Overview and wanted to get stuck right in. I chose one of my past blog posts, imported the URL into a new notebook and then chose the Deep Dive Conversation option under Audio Overview.
A few minutes after clicking on the Generate button I got a wav file which took the form of an AI-generated audio discussion between two AI hosts based on the blog post I gave it. You can, of course, upload a PDF or Google Docs file, use YouTube links, cut and paste text… and more, into the Upload box if you don’t have a blog.
I was completely blown away by the result. It wasn’t a transcript but a full-on conversation between two ‘people’. It sounded incredibly realistic, if but a tad cheesy, and had a laid-back casual feel of a podcast. The discussion expanded on the content I gave it and turned it into an impressive dialogue about the subject. Also, it didn’t get sidetracked with information sources from outside the details I fed it.
In practice, I tested the feature a few times before I was satisfied. This was partly due to my inexperience with the tool but I also found that I had to point out one or two things before it improved upon it. You can give it direction before generating a discussion by clicking the Customise button and adding information there. For instance, I told it my name and an approximate time limit for the file.
I didn’t come across any inaccuracies but there were one or two parts which I would have altered slightly, if I could have. This, unfortunately, is not possible to do at the time of writing – you’d need to regenerate a new discussion and see what you get the next time. Another preference of mine would have been to have a British accent for the characters but, again this isn’t currently available. Finally, I found that sometimes the speech goes a little fast in places and might be difficult to follow for those who are hard of hearing.
The audio file can then be downloaded or shared. After downloading to my PC, I chose to import the wav file into Audacity, which is a free and open-source digital audio editor and recording software program. I then exported the file as an mp3 in order to reduce the file size before uploading to my blog. You don’t need to take this step if you’re happy with the wav format.
You need to be logged in to a Google account to take advantage of NotebookLM and there are limitations with the free version. In my experience I seemed to have three opportunities to create conversations before I was informed that I had to try again the next day. A workaround to this is to use another Google account if you have one and then you can create another Deep Dive Conversation in the second account. Nevertheless, if you’re adding lots of different sources to a particular notebook this might not be practical.
LISTEN
And now for the fun part. To listen to my first AI Deep Dive Conversation, go to my blog post entitled ‘A 1921 Census Proves Tricky In Locating Hibbitt Relatives’ and click the play button on the audio player. More examples can be found in the Audio Enhanced Posts category.
Also, don’t be shy, leave a comment to let me know what you make of it.
