Is it underground velvet or velvet sunset?
Fictional rock band Velvet Sundown has emerged with its music, lyrics, and artwork all generated by AI. This phenomenon sparks discussions about how these technologies blur the boundaries between authentic creation and synthetic output in the music world, raising questions about whether artists should be open about their use of AI.
Music production has long relied on computer software, and now, artificial intelligence has become a critical tool for producers and artists. Platforms like Suno and Udio are revolutionizing how songs are made, allowing users to generate music with just a few prompts.
While some listeners are indifferent to AI-generated tunes, others are keen to know the origins of the music they hear.
If you’ve stumbled upon a new track and wonder if it was entirely created by AI, there are a few methods to uncover its background.
I’ll check the background
If you’re curious about the creators behind a song, a little detective work might help.
According to Manuel Musalam, who heads research at the streaming service Deezer, the most telling signs often come from “external factors.”
Do the artists have social media presence? A silence in the online world might suggest they’re not real. If they do have accounts, check the kind of content they share and their activity level.
Are there any signs that these artists exist outside of the digital realm? Are concerts scheduled? Can tickets be purchased? Is there footage from prior shows available online? Has a known record label released their music?
Looking for the source is also helpful. Creators might publish songs using AI tools like Suno or Udio, though this isn’t always the case. Accessing this information usually requires signing up for an account to explore genres and playlists based on song titles or artist handles. However, tracking down a song can still be tricky without knowing its name or creator.
Song tag
In an effort for transparency, Deezer has marked AI-generated albums to help combat fraudulent activity aimed at quick royalty earnings. The app and website display an “AI Generated Content” label, indicating which tracks were produced with AI tools.
The company’s CEO mentioned that its system uses proprietary tech to identify patterns found in audio made by AI generators. While Deezer hasn’t disclosed how many songs have been tagged since the feature’s introduction in June, they estimate that as much as 18% of daily song uploads are AI-generated.
Song scanner
There are various third-party services that can analyze whether a track is human-made or AI-generated.
I tested some AI-generated songs using a tool from IRCAM Amplification, a subsidiary of a French music research institute. It suggested that the likelihood of being AI-generated ranged between 81.8% and 98%, specifically noting they were made with Suno.
To validate the findings, I also uploaded some older MP3s, which returned a significantly lower probability of being AI-created.
A limitation of IRCAM’s tool is that users can’t link songs directly from streaming services for analysis, which can be a hurdle.
There are numerous other platforms where you can upload files or paste Spotify links, though some of these produce uncertain results or incorrectly classify songs.
Please check the lyrics
AI tools can also aid in examining song lyrics. While serious users often prefer to compose their own words, many casual users may choose to input lyrics directly into the machine. However, AI-generated lyrics frequently exhibit a poor rhyme scheme or repetitive patterns, which might suggest artificial creation. Still, it’s a subjective evaluation.
Some users have noticed that Suno often employs certain words in its lyrics, like “Neon,” “Shadows,” and “Whispers.” Lucas Rams from the Philadelphia area, who has created three albums with his AI band Sleeping With Wolves, noted that he likes to include “neon” in his lyrics. “I don’t know why. I love putting neon in everything,” he said.
There’s no simple answer
With AI technology evolving rapidly, determining the authenticity of content is becoming increasingly complex. Experts suggest relying on ears alone isn’t enough.
