AI music detection is a challenge! In response to comments by readers from the last two posts on this subject, I spent the last ten days researching software tools that are available to assist radio stations and DJs in determining if a song is AI generated. It was disheartening! There are currently no “off the shelf” solutions for individual DJs to use that I liked. There are several solutions for radio stations and other corporate users but their rate of detection still needs improvement. Many of the software options also were based on minutes of use which is impractical and would be expensive given the shear amount of new artists music that many DJs at BMFR are reviewing each month.
So how is the AI Bot Music challenge being met? Some of the major international record companies (e.g. Sony Music, Universal Music, Warner Brothers) have filed copyright infringement law suits against two of the biggest AI music bot generators: Suno and Udio. Essentially the music companies are suing them for the practice of “scraping” the internet and using copyrighted material to train their AI voice, lyrics, and music products.
Two companies that feature numerous AI music bot videos/songs in their content are Youtube and Spotify. Youtube declared last fall that it would make a tool available to help artists search music on the platform for copyrighted content. Spotify does not ban AI music but has stated it will assist in removing content that violates copyright law. In other words, they won’t do much until a lawyer shows up from an artist.
As you have read, most of the response is directed at copyright violations and not the core issue of “is this a song produced using AI generated lyrics, voices, and/or music?” So for now, DJs and radio stations are basically on their own.
So what are the characteristics of a song with AI content and how do you spot them? The voices have a nice but unnatural sound. Now that I am aware of the issue, I find them easy to spot. (I went back through a couple hundred submittals from the last year and found two albums that I initially accepted but now know better!) AI music detection is harder. You are looking for unnatural transitions between notes, weird instrument combinations, and rhythms and harmonies that sound “too good” to be true. And I know of nothing that can help us spot AI lyrics in a song other than a copyright tool.
Some people I have talked to are using AI to spot AI bot music! I was told that ChatGPT will review a song and offer a percentage answer (i.e this song has a 88 per cent likelihood of being AI generated). GROK also offers a likelihood answer for an artist being AI. Statistically, any answer below a 95% level of confidence is just a “coin flip” so again we are on our own! Submithub offers a tool to detect AI music but admits that currently a lot of Suno AI music still qualifies as human. They are working the issue as fast as they can.
The bottom line is that to identify AI music will require DJs and radio stations to do some extra work and use a lot of common sense. The only sure way to combat AI music bots making it on to your playlist is to do research on a new artist before you spin the tune! After hours of going through Youtube, it is easy to spot the AI bot music. They have no links to a social media page, a live video, or listed live dates. Most rely on dozens of hashtag references to well known artists in attempt to gain credibility. And they use only AI generated graphics for the fake artist photos.
Over the next year, a number of software tools are supposed to become available and others will get better. If you know of any software tools that specifically address whether a song is AI generated, please comment below. I will revisit this issue at the end of the summer and let you know if there are any useful updates.
#blues #bluesmusic #bluesnews #bluesreviews #musicnews #newmusic #AI #AImusic
Ben Vee started out spinning songs on terrestrial radio and at nightclubs back in the 1970’s in his home state of Louisiana. After a career in the construction business, he returned to DJing in 2011. He now hosts two shows each week on http://www.bluesmusicfan.com and writes about the blues for BMFR at http:/www.bluesbeats.com and on his own blog at http://www.benveeblues.com from his home in Connecticut.
Thank you Ben, for spot lighting a very frustrating trend in music. I don’t consider AI “artists” as musicians, and therefor, I do not want to support them as legitimate. My biggest sticking point would be the deceptive nature. As you mentioned, they rarely offer up biographies, real photos and of course never play live. Yet, they do not often offer clear disclaimers that their music is indeed AI generated. If they were honest by offering disclaimers, it would soften my view to a degree, but I still would not support them in comparison to real musicians.