How to tell if a song was made by AI or a human

Euronews

5 days ago

Is it The Velvet Underground or Velvet Sundown?
The fictitious rock group, Velvet Sundown – which comes complete with music, lyrics and album art generated by artificial intelligence (A) – is stoking debate about how the new technology is blurring the line between the real and synthetic in the music industry, and whether creators should be transparent with their audience.
Computer software is widely used in music production, and AI is just the latest tool that disc jockeys, music producers and others have added to their production pipeline. But the rise of AI song generators such as Suno and Udio is set to transform the industry because they allow anyone to create songs with just a few prompts.
While some people do not care whether they're listening to AI-generated music, others might be curious to know.
If you encounter a new song that leaves you wondering whether it's 100 per cent made with AI, there are some methods that could reveal how it was created.
Do a background check
If you're wondering who's behind a song, try some old-fashioned detective work.
The 'most obvious cues' come from 'external factors,' said Manuel Mousallam, head of research and development at streaming service Deezer.
Does the band or artist have social media accounts? Lack of a social presence might indicate there's no one there. If they do exist online, examine the kind of content they post, and how long it goes back.
Is there any sign that the artist or band exists in real life? Are there any upcoming concerts and can you buy a ticket for a gig? Is there footage of past concerts on YouTube? Has an established record label released their singles or albums?
Try going to the source. Song creators often – but not always – publish their generated tunes on the Suno or Udio platforms, where they can be found by other users.
The catch is that you'll have to sign up for an account to get access. Users can look up songs by track name or the creator's handle, and browse genres and playlists. But it can still be difficult to spot a song, especially if you don't know the name of the song or creator.
Song tags
Deezer has been flagging albums containing AI-generated songs, as part of its efforts to be more transparent as it battles streaming fraudsters looking to make quick money through royalty payments.
The Deezer app and website will notify listeners with an on-screen label – 'AI-generated content' – to point out that some tracks on an album were created with song generators.
The company's CEO says the system relies on in-house technology to detect subtle but recognisable patterns found in all audio created by AI song generators.
The company hasn't specified how many songs it has tagged since it rolled out the feature in June, but says up to 18 per cent of songs uploaded to its platform each day are AI-generated.
Check the lyrics
AI song tools can churn out both music and lyrics. Many serious users like to write their own words and plug them in because they've discovered that AI-generated lyrics tend to be bad.
Casual users, though, might prefer to just let the machine write them. Bad rhyming schemes or repetitive lyrical structures might be a clue that a song is not man-made, but it's subjective.
Some users report that Suno tends to use certain words in its lyrics like 'neon," 'shadows' or 'whispers'.
If a song includes these words, it's 'a dead giveaway' that it's AI, said Lukas Rams, a Philadelphia-area resident who has used Suno to create three albums for his AI band Sleeping with Wolves.
'I don't know why, it loves to put 'neon' in everything'.
No easy answers
AI technology is improving so quickly that there's no foolproof way to determine if content is real or not, and experts say you can't just rely on your ear.
'In general, it can be difficult to tell if a track is AI-generated just from listening, and it's only becoming more challenging as the technology gets increasingly advanced,' said Mousallam of Deezer.
'Generative models such as Suno and Udio are constantly changing, meaning that old identifiers – such as vocals having a distinctive reverb – are not necessarily valid anymore'.