'AdBlock Radio’ wants to use A.I. and Shazam-like tech to end podcast ads

Its goal is to play a role in creating 'new experiences for streaming media' worth paying for.

Wellington Nem listens to Apple Airpods.

No matter where you go, ads in our real and digital lives are incessant. And in the opinion of Alexandre Storelli, creator of a tool called ‘Adblock Radio’, they’re not only a nuisance but a dangerous way companies are manipulating their audience.

“Ads exploit the weaknesses of many defenseless souls,” Storelli told Motherboard Wednesday. “Ads dishonestly tempt people, steal their time and promise them a higher social status. Blocking them will be a relieving experience for many.”

In his interview with Motherboard, Storelli states that the over-consumerism encourages by this onslaught of ads is contributing to the climate crisis facing our planet. And while that itself may be a bit of a leap, on his blog, Storelli continues on to say that he’s not necessarily advocating for a totally free audio world — after all, podcasters and radio hosts need to make a living somehow — but that he instead advocates for a shifting of the business model that would support “new experiences for streaming media that will be worth paying for.”

Radio: a big holdout from the ad blocker.
Radio: a big holdout from the ad blocker.

Adblocking plug-ins for your browser are already a common remedy for web or video-based ads, but when it comes to audio the challenge of blocking them becomes a little trickier because you first need a tool that can listen to and identify an ad from music or conversation. Storelli has been at this project for nearly four years now and describes a trial and error process which included failed approaches like speech recognition, looking for patterns in volume level or timing, and crowd sourcing. While some had marginal success, they were far from reliable.

Storelli only more recently settled on an approach that can reliably identify and block these ads for listeners, using machine learning to sort types of audio and something called ‘acoustic fingerprinting’ to refine those sorted results.

For Adblock Radio, differentiating between music, talking or advertisement is essential. To do this, Storelli started by training a neural network on small samples of audio that had sections identified as either music, advertisement or talking to help it learn to identify the different characteristics of these types. By November 2018, Storelli says he’d trained the network on “66 hours of ads, 96 of talk and 73 of music.”

If the network identifies an audio sample incorrectly, Storelli says that users can flag the error in Adblock Radio. But, Storelli found that the neural network itself was not enough to power Adblock Radio.

Storelli says that by 2018 Adblock's neural network had been trained on "66 hours of ads, 96 of talk and 73 of music" to recognize and categorize different types of audio.
Storelli says that by 2018 Adblock's neural network had been trained on "66 hours of ads, 96 of talk and 73 of music" to recognize and categorize different types of audio.

To further refine the categorizations made by the neural network, Adblock also uses a technique called acoustic fingerprinting, which is more or less what it sounds like. Audio fingerprints are created by converting audio features into a series of numbers. Known ads are fingerprinted and stored in a database. If a live ad is played that matches something in the database, it is identified and blocked.

This is a similar approach to that of the music-identifying service Shazam, which was originally a stand-alone app before being bought two years ago by Apple and integrated into its music service and Siri. The main difference is that instead of combining it with machine learning like Adblock Radio, Shazam sticks mostly to using only acoustic fingerprinting. This is likely because, as a music-specific service, it has little need to also differentiate between talking as well.

An obvious downside to Adblock Radio is that acoustic fingerprinting is restricted to identifying only ads it’s heard before. Storelli says he’s looking into ways to help automate the database maintenance process as well.

Together in harmony, Adblock Radio’s neural network identifies types of audio and the acoustic fingerprint database helps verify it if the network gets confused. Like any good digital crusader, the code for Adblock Radio is available for free on GitHub.

Adblock Radio is currently compatible with 79 radio stations across 13 countries and is looking to expand. But, while an ad blocker like this may seem innocuous to users, it is a big concern for media companies and one that companies like Facebook has already taken action to push-back against when it comes to web ads.

Instead of bankrupting these companies, Storelli hopes that Adblock Radio will force them to find new and more conscientious business strategies that consider the effects of the products they promote and the methods they use.

“Adblock Radio’s long-term vision is to create new experiences for streaming media that will be worth paying for,” Storelli wrote on his blog. “[S]o that artists get paid and ads won’t be necessary anymore.”