The 125th Block: An impractical way to spot deepfake audio

Also, why platforms are boring and memes are radical

Sep 25, 2022

This week…

I’m looking to find the time to watch some Jean-Luc Godard films. I enjoyed Breathless (1960) and might rewatch it. The French filmmaker died earlier this month from assisted suicide. Tell me your favourite Godard feature film—and please, spare me from the more ~~bizarre~~ highbrow ones (iykyk).

And now, a selection of top stories on my radar, a few personal recommendations, and the chart of the week.

Deepfake audio has a tell – researchers use fluid dynamics to spot artificial imposter voices

Deepfaked audio often results in vocal tract reconstructions that resemble drinking straws rather than biological vocal tracts. Logan Blue et al., CC BY-ND

Logan Blue and Patrick Traynor on The Conversation:

The first step in differentiating speech produced by humans from speech generated by deepfakes is understanding how to acoustically model the vocal tract. Luckily scientists have techniques to estimate what someone – or some being such as a dinosaur – would sound like based on anatomical measurements of its vocal tract.
We did the reverse. By inverting many of these same techniques, we were able to extract an approximation of a speaker’s vocal tract during a segment of speech. This allowed us to effectively peer into the anatomy of the speaker who created the audio sample.
From here, we hypothesized that deepfake audio samples would fail to be constrained by the same anatomical limitations humans have. In other words, the analysis of deepfaked audio samples simulated vocal tract shapes that do not exist in people.
Our testing results not only confirmed our hypothesis but revealed something interesting. When extracting vocal tract estimations from deepfake audio, we found that the estimations were often comically incorrect. For instance, it was common for deepfake audio to result in vocal tracts with the same relative diameter and consistency as a drinking straw, in contrast to human vocal tracts, which are much wider and more variable in shape.

Well, then, I would like to know how to put this to practical use.

How platforms turn boring

Russell Brandom for The Verge:

TikTok is changing.
Even a year ago, my For You page was mostly stuff you could only see on TikTok, whether it was Vine refugees making comedy shorts or song memes like Here Comes The Boy. That stuff is still on the platform, but it’s largely fallen off my For You page, replaced by Tim Robinson sketches and funny animal videos. One account reposts Derry Girls clips with captions about the royal family; another (aptly named ViralHog) licenses viral video clips from local news or Reddit threads and then blasts them out to different platforms. Everything has the same warmed-over feel. For many of these accounts, the goal is to churn through enough content to build a following so they can flip the account into advertising mode for a quick buck.
Of course, since it’s mostly content that has already blown up elsewhere, there’s usually something compelling about it. It’s not bad content really, but it’s an ominous sign for the platform. At first, TikTok was exciting because there was culture that could only happen there. Now that on-platform culture is being overwhelmed by viral arbitrage, and the actual content is getting closer to what you see on every other network. As the platform gets bigger, it gets more generic, and there’s less to distinguish it from every other mass-market social network.

The number of manual configurations (through muted words, accounts, etc.) I do to curate my social media feeds to no avail tells me that users are at the mercy of platform owners. Plus, Mozilla reports that YouTube’s dislike button is barely effective (see below).

Does this button work? Investigating YouTube’s ineffective user controls

Becca Ricks and Jesse McCrosky for Mozilla Foundation:

To evaluate the effectiveness of YouTube’s controls for people who use the platform, we carried out a study that leverages Mozilla’s large community of RegretsReporter volunteers. 22,722 people donated their data to Mozilla, generating a dataset of the 567,880,195 videos they were recommended. This study represents the largest experimental audit of YouTube by independent researchers, powered by crowdsourced data.
To understand whether people feel in control, Mozilla surveyed 2,757 RegretsReporter participants to better understand their experiences with YouTube’s recommendation algorithm.
Ultimately we wanted to learn whether people feel in control on YouTube – and whether those experiences are actually validated by our RegretsReporter data. By combining quantitative and qualitative insights in this research project, we aim to paint a more complete picture of how YouTube’s recommendation algorithm handles user feedback.

TL;DR: YouTube’s user controls don’t work for most.

How memes led to an insurrection ($)

Joan Donovan, Emily Dreyfuss, and Brian Friedberg for The Atlantic:

“We’re storming the Capitol! It’s a revolution!” a woman who identified herself only as Elizabeth from Knoxville, Tennessee, told a reporter outside the U.S. Capitol on January 6, 2021. She had a blue Trump flag slung across her neck like a cape. Police had just maced her in the face, she recounted through tears.
As soon as this video of Elizabeth hit Twitter, it went viral. The insurrection she’d taken part in, a very real and coordinated attempt to thwart the democratic process, was also a surreal spectacle—millions watched the chaos unfold in real time on broadcast TV and on social media—and Elizabeth was one of the minor characters. On Twitter and TikTok, she became fodder for internet jokes. People remixed the video with Auto-Tune. Sleuths spun conspiracy theories: Maybe Elizabeth was a liar who hadn’t really been maced. Maybe the insurrection was a hoax. Content was crafted to fit different political orientations and different platforms, and to delight or offend different audiences.
In short: Elizabeth from Knoxville had been memed. No longer a person with a real identity, now she was a memorable piece of media. The video clip of her was recontextualized, remixed, and redistributed, carrying all sorts of meaning. That’s the definition of a meme: a resonant, authorless idea that spreads through culture, evolving with every hand that touches it.

What I read, listen, and watch…

I’m reading “Take Care of Your Blog” by Robin Rendle. “Ignore the analytics and the retweets though. There will be lonely, barren years of no one looking at your work. There will be blog posts that you adore that no one reads and there’ll be blog posts you spit out in ten minutes that take the internet by storm.” Aye.
I’m listening to Can I Tell You a Secret, hosted by The Guardian journalist Sirin Kale who investigates the story of Matthew Hardy and his decade of cyberstalking.
I’m watching What We Do in the Shadows, the TV series. It’s funnier than it should be.