Google just launched its most impressive (and unsettling) addition to AI Overview yet, a new feature called “Audio Overview” that generates audio summaries of search results, narrated in the style of two life-like, yet not quite human podcast hosts.
Audio Overview is currently an opt-in Search Labs feature, meaning you won’t see the option for it unless you toggle a switch in Search Labs. Right now it’s only available in the U.S. and only generates English summaries.
I tried out Audio Overview myself and the results weren’t exactly what I was expecting.
After you activate the feature in Search Labs, some Google searches will include an Audio Overview box, usually below the regular AI Overview and “People also ask” sections. You just tap the button to generate the audio summary and wait for it to finish processing.
The audio clip you’ll get is generated on the spot, so if you refresh the page and generate it again, it could end up being different. The summaries I generated ranged from 3 to 5 minutes long. All of them feature the same pair of AI-generated voices who go back and forth discussing whatever topic you searched, in the style of a podcast.
The voices are admittedly significantly more lifelike than the robotic Siri sound I was expecting. There’s tone changes, conversational word choices, seemingly natural language. It’s not quite realistic, though. The two voices are like a pair of podcasters with zero rapport who seem like they’re reading off a teleprompter. It’s just shy of being lifelike but still lifelike enough that some people could be fooled at first.
Google shows you which search results it used to generate the audio summary, so you can double-check whatever info your AI podcasters give you. However, they sound realistic enough that some people might just assume these are real people and take whatever they say as fact. Of course, that’s also an issue with text AI summaries.
There are some hiccups that give away that these aren’t real people. For instance, in one of the summaries I got, one of the AI voices asks a question then immediately answers it herself, which sounded pretty awkward.
Both voices use emotional language from time to time, like exclaiming “Wow!” at a fun fact, but it definitely sounds stiff and just shy of something a real person would say. The AI voices also mispronounce words once in a while, like “musk” instead of “must.”
Uncanny and eerie as this feature is, I can see it being helpful to some people, especially those who may have vision impairments, or otherwise rely on screen reading tools. The AI-generated voices sound pretty good for the fact that they’re AI, too.
That would be cool if it didn’t come with a host of concerns around the spread of misinformation through AI and the threat AI-generated voices like this could pose to jobs like voice acting. Like any innovation in AI, Google’s Audio Overviews are a double-edged sword and unfortunately I’m still more skeptical than impressed.