We know that generative AI models are now smart enough to produce images, videos, text, and audio that—most of the time—can convincingly pass for something created by a human. These types of content can be produced in seconds for little or no money, which means we’re seeing them hit the internet and its various platforms en masse.
While there are protections in place to stop impersonation and spam, AI-made music isn’t banned on the big audio streaming platforms. In some cases, it’s proved as popular as music made by humans without any AI assistance. But what if you want to make sure that everything you hear is authentically human?
The music streaming service Deezer has now developed a sophisticated AI song detector you can use for free, across 20 different platforms and for up to 100 playlists—you don’t even need a Deezer account to use it. Here’s how it works, and what happened when we put it to the test on a few playlists.
Introducing the Deezer AI music detector
Defining AI music can be tricky: Many artists use various kinds of software to enhance their music, produce synthetic sounds, and fill in some of the gaps left by human musicians. With that in mind, the line between AI-free music and music made by machines can be a little blurry.
Deezer says its detection tool is set up to spot music that’s fully AI-generated, and that it has a success rate of 99.8 percent. In contrast, its research shows that 97 percent of its listeners can’t tell the difference between AI-made music and human-made music when given a blind test.
The idea of the tool is not to ban AI music from Deezer, but to accurately label it so that listeners know what kinds of tunes they’re adding to their playlists. According to Deezer, 43 percent of people who switch to the platform from an alternative like Spotify or Tidal and bring their playlists across have some AI tracks in their library.
Understandably, Deezer doesn’t reveal exactly how it detects AI music to stop people finding ways around it. What it does say is that “generative software leaves specific artifacts in the audio signals of AI music,” and judging by the accuracy rate that Deezer is claiming, it works well.
Another stat from Deezer: 44 percent of the tracks uploaded to the service every day, around 75,000 songs, are AI-generated. These AI tracks don’t show up in Deezer recommendations or editorially curated playlists and account for only 1-3 percent of streams on Deezer—with up to 85 percent of those streams identified as fraudulently manipulated in some way.
Putting it to the test
You can try the free Deezer AI detector tool in your browser (available here): Just click Scan my playlists and follow the instructions on screen. Take your pick from the list of streaming services this works with (which is pretty much all of them), then enter the relevant login details so that Deezer can gain access.
Deezer says the only data it gets from your account is securely protected, and limited to what it needs to analyze your playlists—so no contact or payment information, for example. You do have to give Deezer’s AI detection tool permission to connect to your account, but you can revoke access afterward through the settings in your music streaming app.
I tried this with my Spotify library first, because as far as I’m aware, there isn’t any AI music in any of my playlists there. After a few minutes of importing and analyzing, my score came back as 0 percent—my Spotify account is free of AI, as expected. On the results screen, you get the option to share your score, and of course sign up to Deezer if you haven’t already.
Next, I gave the AI detection tool a go at my YouTube Music account, which includes some ambient, lo-fi mixes that I think have been generated by AI. This time, the scanner did detect a trace of AI: 1 percent, which is probably about right considering the size and number of my playlists. However, Deezer won’t tell you which songs are actually AI-generated.
AI music clearly isn’t going anywhere, but for a lot of listeners, it’s important that it’s labeled as such—so that it can be avoided or not, as people prefer. It seems that this type of AI creation might be a little easier to detect than writing or imagery, which will be useful as the deluge of artificial tunes continues.
