Why Podcasts?
While podcasts can’t replace teaching and learning in the classroom, they can certainly enhance the learning experience of your students outside of class. This has been recognized by many including WCUPA faculty member Dr. Anne Gill (see this episode of the ODLI On Air titled “Podcasts as a Course Text“). By providing an audio recording of content related to material covered in the classroom, you’re providing students with a resource that is accessible at all hours and caters to diverse learning styles and schedules. Additionally, podcasts are generally more conversational than traditional educational media, which can create a sense of connection and intimacy between the listener, the hosts, and the content. Providing podcasts to students also provides opportunities to engage with content in more situations and contexts than traditional educational media. The vast majority of podcast listening occurs while doing the dishes, mowing the lawn, while commuting, or while performing other tasks that don’t necessarily require your sense of hearing. Audio podcasts, then, naturally invite engagement with their topics in more situations than would reading a textbook or watching a video recording. If we’re lucky, our favorite podcasters will have a discussion of one or multiple topics that we have covered in class. In these situations, instructors can recommend or assign the podcast episode to the students. But, what can we do if we’d like to engage students with a podcast that no one has made yet? Well, you could use the power of generative AI to create an engaging podcast between two AI voices on any material you provide using a tool like Google’s NotebookLM.
What is NotebookLM?
NotebookLM (originally announced as Project Tailwind in 2023) is a free, previously experimental tool from Google that was made in collaboration with Google’s AI engineers and renowned author, Steven Johnson. Together, they created an AI-enabled tool capable of leveraging the power of large language models to ask and answer questions. At first glance, this might seem like any other chatbot in existence (ChatGPT, Gemini, Claude, etc.). But, this collaboration sought to narrow the scope of AI responses to articles, recordings, and other sources of truth uploaded by the user as well as providing an easy interface to cite the specific sources used to generate text. In other words, NotebookLM was designed as a tool to not only help humans process vast amounts of data from disparate resources, it also provides an interface that allows humans to fact-check AI responses.
But what about AI Podcasts?
While podcast generation was not an initial goal of NotebookLM, the “audio overview” option, added in September 2024, has become a popular tool among users. With a single click of a button, a user can create a podcast with two different AI-generated voices discussing and summarizing sources uploaded to NotebookLM. Most importantly, these two voices take on different personas and mimic speech patterns and intonation that convincingly resemble authentic human conversation. Unlike past versions of text-to-speech audio assistants (e.g., Alexa, Siri, and Google Assistant, etc.), these voices appear less-robotic due to the introduction of so-called disfluencies: pauses in speech, filler words (e.g., “uhm” or “so”), introduction of meaningful vocalizations (e.g. “mmmhmmm”), and other natural unconscious patterns of speech that you would expect in real-life conversation. The illusion is so convincing that I often feel like these robotic voices are more human than recordings of real humans in audiobooks (no human naturally speaks the way that humans write). The end result of NotebookLM’s ability to accurately synthesize information combined with realistic voices is a roughly 10 minute audio file that discusses content with humor, banter, and back-and-forth dialogue; all on any topic for which you provide sources. Here, as an example, is one of the AI podcasts I generated for my Human Anatomy and Physiology students based on a lecture I gave on epithelial tissue.
Step-by-step guide to generating podcasts
So how do you go about making on of these AI podcasts for your students?
Collecting Sources
Perhaps the most difficult part of this process involves deciding and/or capturing sources to feed into NotebookLM. If you want to upload lecture notes or a slide deck, you can upload them as PDFs or plain text files. At the time of writing, the following files can be uploaded: PDF, .txt, Markdown, and audio (MP3/wav). As a general rule-of-thumb, the inclusion of more sources generates longer audio files. Personally, I like to generate podcasts using a single source – recorded audio from my synchronous in-person lectures. I’ve found that this tends to result in the AI podcasters referencing specific material and examples I cover in class and provides my students with reinforcement. I use a standalone external audio recorder to capture my audio (DJI mic 2), but you could easily use a voice memo app built into your mobile device or computer. Once you’ve decided on what sources you’d like referenced in your podcast, the next step is to create a notebook and upload the files.
Creating a Notebook
To create an AI-cast, you’ll first need to create a notebook. Here is a video recording explaining how to do this process.
In summary, here are steps to create your AI podcast:
- Go to http://notebooklm.google
- Click “Try NotebookLM”
- Click “+ Create new”
- Drag your sources into the upload window
- Click Generate
Easy Personalization
If you want to customize your podcast a little, you can click the “Customize” button instead of the “Generate” button. You won’t have this option if you’ve already generated audio, so you’ll need to delete the generated audio using the 3 stacked vertical dots first. Clicking on the “customize” button allows you to provide the AI generation some prompting to have some input on what the AI hosts talk about or at what level they talk. Here is a prompt I often use for my class:
“This class is bio 259 taught by Dr. Mike Rosario. It is taught at the college level. You are two students in the class that decided to make a podcast about this lecture. Also, throughout, ask clarifying questions. Try to reference specifics from the lecture and ask each other to elaborate. Use specific examples from class when possible.”
After inputting your prompt, click “Generate” and wait for your new podcast to become available.
More-involved Personalization
If you want to put yourself in the conversation, an update on December 13, 2024 allows you to do so! After generating an audio podcast, the new “Interactive mode” lets your voice be part of the conversation. Enabling interactive mode will provide a slightly different UI. At first, it will seem no different from the previous examples, except there’s a new button that you can click named “Join.” Clicking this button will interrupt the AI hosts, and, if you give your web browser microphone permissions, NotebookLM will bring you into the conversation so you can contribute and guide the discussion. NotebookLM does not currently save your voice files or the AI hosts’ responses, but if you record yourself and the hosts using recording software (e.g., Audacity), you can capture the audio for sharing. Here is an example of me interacting with the AI hosts (with a little bit of audio and video editing) about a lecture I have on reflexes in the human body.
Conclusion
This new ability to seamlessly integrate your own voice into NotebookLM’s AI-generated conversations opens up a whole new world of possibilities for engaging your students. Imagine creating a podcast where you act as a moderator, prompting AI hosts to explore different aspects of a topic, or even having students take turns joining the conversation themselves!
By utilizing NotebookLM’s capabilities, you can create a dynamic and interactive learning experience that goes beyond the traditional lecture format. Let your creativity flow, experiment with different prompts and techniques, and most importantly, have fun exploring this exciting new tool for enhancing your classroom!
