After the tool generates an AI podcast, you can create a shareable link to the audio or simply download the file. You also have the option to adjust the playback speed in case you want the podcast to be faster or slower.
The Future of AI Podcasts
The Internet has gotten creative with NotebookLM’s audio feature, using it to create audio-based “deep dives” into complex technical topics, generate files that neatly summarize dense research papers, and create “podcasts” about their personal health and fitness routines. Which raises an important question: I have to using NotebookLM to view your most personal files?
The summaries generated by NotebookLM are, according to Google spokesperson Justin Burr, “based entirely on the source material that the user uploads. This means that your personal data is not used for NotebookLM training, so any personal or sensitive information you have in your sources will remain private unless you choose to share your sources with collaborators.” So far, this appears to be a of the benefits of Google labeling NotebookLM “experimental”; to hear Google’s frame of it, the company is just gathering product feedback right now, being flexible and responsive, tinkering in a lab, and NotebookLM is separate from its multibillion-dollar ad business. For now! For now.
Adding audio options to Google Labs’ online notebook was a transformative moment. “By changing the modality, it unlocks a whole new set of use cases,” says Martin. What makes NotebookLM stand out from all the other generative AI tools available to users in 2024 are, surprisingly enough, the extra words and odd phrases. Instead of the drab, monotone voice-over you might expect from two AI voices summarizing data, the rhythm and vocal performances of NotebookLM’s synthetic podcasters sound far less bloated.
Should podcasters be shivering in their soundproof booths right now? Not quite. Even if podcasting AI tools like the one in NotebookLM turn out to be sticky and engaging summaries of information for the general public, which remains to be seen, synthetic voices will never fully mimic the parasocial bonds developed by human podcasters talking nonsense for hours on end. their subscribers voyeuristically listen.
These audio reviews are not meant to match a specific podcaster voice, mind you, but rather a kind of idealized podcaster duo. Easily recognizable by their “ums,” “ohs,” and free-flowing, pausing conversational style. “Even just from the first week we launched, it was clear what the roadmap was after that,” says Martin. “People want the knobs.” Allowing users to further modify the AI’s output, such as the length of the podcast or the topic of focus, is a priority for the team, and she hopes to ship updates quickly.
Adding more languages and different accents is also important to her. Synthetic hosts are currently calibrated for English-only conversations. However, don’t expect to be able to use your own voice in the NotebookLM podcast generations anytime soon. Martin says the team needs to see if this is a feature people really want and if it can be implemented responsibly.
The explosive popularity of NotebookLM’s Audio Overviews as part of Google Labs, rather than as a feature in Gemini’s chatbot, is a reminder that AI companies aren’t entirely sure what will resonate with users until the software is out in the wild. OpenAI’s ChatGPT was originally released as a preview, for example. And within the constant flurry of generative AI announcements, whatever captures the zeitgeist isn’t necessarily the most marketable or utilitarian feature, but rather the most fun.
Additional reporting by Lauren Good.
