Why our AI Music Detector stays ahead?
Our AI Music Detector leads the industry in identifying AI-generated tracks, ensuring artists are protected...
Over the past 6 months, we have been hard at work to stay as close as possible to market trends. Our AI Music Detection tool has been benchmarked by independent firms, tested by many of our users, and acquired by a significant number of customers.
We thought it was time we gave you insights into what is today considered the most ground-breaking innovation to safeguard your listeners and customers against AI-generated music.
Generative music isn’t an issue per se, unless it shatters the royalty pool and overshadows upcoming artists, lowering their chances of success.
In order to make our detection tool as robust and aligned with market trends as possible, we have gathered a team of field experts to highlight which models were the most concerning to them. This selection contains both commercial platforms and open-source models. This prioritization framework allowed Ircam Amplify to focus on what really matters: discovering and filtering out 80% of the suspicious, poor-value AI-generated content.
And this is how we decided to focus on:
👉 Read our development diary here: Our AI-Generated Music Identification journey
We take pride in being able to detect not only clean, unedited content generated from these platforms and models, but our in-house AI was also trained to detect highly altered content. Whether it’s been sped up, slowed down, or pitched, we deliver the right verdict.
This approach makes our AI Music Detector one-of-a-kind on the market, ensuring that no content remains undetected and providing true visibility on your catalog or content.
Our data science team has also created a unique framework to ensure no model takes more than 10 days to be added to our detection tool: A new version of SUNO just got released? Already detected. You’re dealing with tracks from a lesser-known model that represents 80% of your AI-generated content? We've got you covered.
There is no model that we can’t add, and chances are, we already detect it today.
Not convinced we’re the best out there? After our various (internal and independent) benchmarks, we thought it would be useful to share some data.
Once a subject of curiosity, AI-generated music has taken the industry by storm. To get an idea, just be aware that users of the music AI app Udio alone are generating an average of 864K new songs every single day. Putting one of these tracks on the major DSPs takes hours and only requires a small flat fee.
On the remuneration front, the pool shared to pay out artists on the biggest streaming platform has already been negatively impacted—for real artists, that is. As insane as it seems, AI “artists” regularly make their way onto Spotify's curated playlists, reaching numbers as high as 240K monthly listeners and half a million streams for their top track.
A recent report on AI and music from the Australasian CMO reveals that 23% of music creators’ revenues will be at risk due to generative AI by 2028, amounting to an estimated cumulative loss of over half a billion AUD (around USD 350 million).
Even worse, AI-generated music actively serves streaming fraudsters. By producing and uploading thousands of tracks to be listened to by bots, fraudsters can generate a small number of streams on each track, keeping them below the radar. Yet, they make incredibly huge numbers by accumulating streams across all tracks, as illustrated by the $10M fraud case currently under investigation, involving a “musician” and… an AI company CEO.
To really address the AI-generated tracks challenge you need to implement identification as part of your quality assurance process—no more, no less. Which means theoretically performing up to 100k scans a day, to cope with the volume of tracks added daily to streaming platforms. If only one service were to ingest and run AI music detection on this amount, our tool’s analysis time would allow it, as we’re able to screen up to 5,000 tracks in less than a minute.
As we’re striving to enable transparency on AI-generated content for the industry as a whole, we want to make it as cost-effective as possible. It should be a no-brainer, no matter the size of the structure. We’re talking about USD cents per track.
Obviously, for larger companies managing the big numbers quoted above, we can discuss corporate rates based on the volumes you’re handling.
Think we're on the same wavelength?