Every day, tens of thousands of songs are released. This constant stream of options makes it difficult for streaming services and radio stations to choose which songs to add to playlists. To find the ones that will resonate with a large audience, these services have used human listeners and artificial intelligence. This approach, however, lingering at a 50% accuracy rate, does not reliably predict if songs will become hits.
Now, researchers in the US have used a comprehensive machine learning technique applied to brain responses and were able to predict hit songs with 97% accuracy.
"By applying machine learning to neurophysiologic data, we could almost perfectly identify hit songs," said Paul Zak, a professor at Claremont Graduate University and senior author of the study published in Frontiers in Artificial Intelligence. "That the neural activity of 33 people can predict if millions of others listened to new songs is quite amazing. Nothing close to this accuracy has ever been shown before."
Study participants were equipped with off-the-shelf sensors, listened to a set of 24 songs, and were asked about their preferences and some demographic data. During the experiment, the scientists measured participants' neurophysiologic responses to the songs. "The brain signals we've collected reflect activity of a brain network associated with mood and energy levels," Zak said. This allowed the researchers to predict market outcomes, including the number of streams of a song – based on the data of few.
[...] After data collection, the researchers used different statistical approaches to assess the predictive accuracy of neurophysiological variables. This allowed for direct comparison of the models. To improve predictive accuracy, they trained a ML model that tested different algorithms to arrive at the highest prediction outcomes.
They found that a linear statistical model identified hit songs at a success rate of 69%. When they applied machine learning to the data they collected, the rate of correctly identified hit songs jumped to 97%. They also applied machine learning to the neural responses to the first minute of the songs. In this case, hits were correctly identified with a success rate of 82%.
"This means that streaming services can readily identify new songs that are likely to be hits for people's playlists more efficiently, making the streaming services' jobs easier and delighting listeners," Zak explained.
[...] Despite the near-perfect prediction results of his team, the researchers pointed to some limitations. For example, they used relatively few songs in their analysis. Furthermore, the demographics of the study participants were moderately diverse, but did not include members of certain ethnic and age groups.
Journal Reference:
Sean H. Merritt, Kevin Gaffuri, Paul J. Zak, Accurately predicting hit songs using neurophysiology and machine learning, Front. Artif. Intell., 20 June 2023, Volume 6 - 2023 | https://doi.org/10.3389/frai.2023.1154663
(Score: 5, Insightful) by Ox0000 on Sunday September 10 2023, @06:06PM (13 children)
Wonderful, this means that that whole industry will devolve even more into a drab, homogenous, and bland melange of things that have been done in the past. You thought Hollywood's sequels and remakes were bad?
All because some executive, somewhere, went: we're making n billion dollars of profit on the backs of artists... You know what would make my life so much better? The thing I really need? An EXTRA m billion of pure profit... THAT would make my life so much better...
Ain't nothing an MBA cannot fuck up...
(Score: 4, Interesting) by khallow on Sunday September 10 2023, @06:20PM
(Score: 5, Interesting) by JoeMerchant on Sunday September 10 2023, @06:55PM (9 children)
> that whole industry will devolve even more into a drab, homogenous, and bland melange
They hit rock bottom decades ago, is it surprising that they continue to dig?
You want something interesting? Compose it yourself. You've got more recording studio in the device you're communicating with this message board than the Beatles ever had at Abbey Road. Synthesized orchestras, basically all that's lacking these days are synthesized vocals. Dylan proved: if you've got something interesting to say, you don't have to be easy on the eyes, or the ears, to have a hit. Same for distribution, you can have an .mp3 on the player of anyone on the planet (who cares to listen) within minutes of releasing your masterpiece(s).
The reason Dylan and friends were so popular is that they resonated with their audiences, messages that mattered to the listeners. That's the trick today: what matters? What are today's listeners willing to get behind, as a group of any size, and stand up and say: "I'm with him/her/zim/hir/them!" ?
🌻🌻🌻 [google.com]
(Score: 2) by krishnoid on Sunday September 10 2023, @08:23PM
Hell, there's probably more processing power in a Javascript-enabled web *page* than the Beatles had. Also, knowing what songs the algorithm identified as "popular" means you can automatedly sequester them, and then look elsewhere for other stuff you might like. Isn't this sort of what Spotify already does anyway?
(Score: 4, Interesting) by Thexalon on Monday September 11 2023, @12:36AM (3 children)
So the mantra in that part of the music world is "three chords and the truth". Before Bob Dylan, Woodie Guthrie really showed how this was done.
But the mantra for pop music is "four chords and meaningless pablum, performed by utterly replaceable 'artists' chosen more for their dancing and looks than musical skill". And musicians are told, more or less, that the price of making it big is to fit that. Which is why Stephani Germanotta, a talented pianist and singer who could belt out jazz and such, had to become the much less interesting Lady Gaga to have a musical career.
"Think of how stupid the average person is. Then realize half of 'em are stupider than that." - George Carlin
(Score: 2) by JoeMerchant on Monday September 11 2023, @01:00PM
Pat Benatar, and many others, also reigned in their musical range to fit the "pop" mold well enough to get promoted.
Four chords and meaningless pablum is the formula proven to sell (but does anybody pay for music anymore?) Occasionally there's a style shift, like Daft Punk or Fluke or The Chemical Brothers that put together an attractive enough image to get significant(?) followings, and of course there are movements like rap that give a group of listeners something to call their own.
Then you get things like Moby... and more similar than not: Nine Inch Nails.
🌻🌻🌻 [google.com]
(Score: 1, Flamebait) by ChrisMaple on Monday September 11 2023, @06:20PM
Woodie Guthrie's audience was leftists hostile to private property, and people who liked the sound of folk music but didn't understand the message. Calling his message "truth" is vicious.
(Score: 2) by gnuman on Tuesday September 12 2023, @09:57PM
4 chord songs :-)
https://www.youtube.com/watch?v=5pidokakU4I [youtube.com]
(Score: 2) by maxwell demon on Monday September 11 2023, @03:29AM (1 child)
Well, that requires one thing that's still not available to everyone: The ability to compose something interesting.
The Tao of math: The numbers you can count are not the real numbers.
(Score: 2) by JoeMerchant on Monday September 11 2023, @01:08PM
>The ability to compose something interesting.
... to a significant number of people. There's the question: at what point is a following significant? If you can fill a bar for 4 hours 3 nights a week, that's starting to be significant.
If I were in the watering hole business, I'd look into starting "zoom jams" where you showcase bands live on the A/V systems that most watering holes have these days. One band could play "live" to dozens or hundreds of bars, get paid enough money to actually work on their music _as_ their day job, and the bar patrons could get decent "live" entertainment - which would really be live at one place and simulcast across the network to other places that like that style. Get a few thousand bars involved, with a hundred artists or less and let them compete "America's Got Talent" style with trivia keypads as the judge inputs. Top talent plays weekends, up and coming on week nights... all you have to do is convince a completely disorganized collection of small business owners to participate without letting them strangle the profits out of the venture.
🌻🌻🌻 [google.com]
(Score: 2) by xorsyst on Monday September 11 2023, @01:44PM (1 child)
No synthesized vocals? Have you not heard of Hatsune Miku?
(Score: 2) by JoeMerchant on Monday September 11 2023, @08:53PM
I suppose it's more accurate to say: the synthesized vocals aren't yet competitive with live singers, for mass market appeal at least.
While guitar (and most other string and wind instrument playing) has near infinite range of possible expression, the synthesizers can already duplicate a large portion of how those instruments are popularly played.
Vocals? the synths can make a few interesting expressive voices, but nowhere near what a half decent singer can.
🌻🌻🌻 [google.com]
(Score: 3, Interesting) by looorg on Sunday September 10 2023, @07:36PM (1 child)
Didn't they put down the formula for a popstar sensation quite a while ago? This is if anything just a refining of the process. Trying to cut down a bit more on the uncertainty factors. But they have it more less down to an art now don't they? Cute boy(s) or a girl, songs and lyrics are almost the same. Combinations of cords that are known to work and sound nice. Massive PR and media machinery plus good looks >> talent.
It's practically risk mitigation. Cause eventually a lot of them appear to implode in social- media due to their ego boost colliding with reality and they say or do something out of this world. A lot of them clearly can't handle the pressure that comes with being a celebrity of today, in some regard I'm not sure anyone can in the long run. So you better strike it rich fast and then be able to ride that until retirement where your fans like you for you or your talent and they don't give a shit what is said or done in the media spotlight.
I wonder if a lot of the rock- and pop-stars of decades long ago would have cut it in today -star market. Even genres that was previously mostly middle-aged or older men mostly are now filled with boys and girls singing arn't they? The age and level of alcoholism have gone down a lot for countrystars etc has it not?
(Score: 3, Insightful) by anubi on Sunday September 10 2023, @11:20PM
This got going really good back when I was a kid...In the 60's.
Remember the "Wrecking Crew"? Probably not. They were the group of professional studio musicians, who created background music for Hollywood movies. They provided most of the instrumentation of Pop Rock groups hit songs.
The performing artists were to look good, dance, gyrate, sing, and do all the social stuff a star does. And they did it well. Auto-tune vocal correction tech gave their voices a full sound, and professional musicians made great sounding instrumental backing so recording studio time wasn't wasted on numerous newbie retakes.
Then Phil Spector. Did he ever make an impact on our music. His "Wall of Sound" concept to marry lots of studio musicians to teen singers.
Here is one of his first hits. It changed everything.
"Be My Baby", the Ronettes, 1963
https://m.youtube.com/watch?v=NnWOBMQhNBQ [youtube.com]
Although I admire Phil Spector's work, he was one sunuvagun to work with, I hear.
"Prove all things; hold fast that which is good." [KJV: I Thessalonians 5:21]
(Score: 4, Insightful) by https on Sunday September 10 2023, @07:26PM (7 children)
Reading through it, it seemed particularly verbose, in a eerily familiar style - SEO spamsite structure, but post-enlightenment phrasing instead of post-modern. It has the smell of ChatGPT or some other LLM. But then these numbers show up, and it's firmly thrown into the trash can. Copied from the article:
... and from a calculator:
No amount of fucking around with significant figures or gender categorization gets you 47%. Options are:
a) it's a bullshit article, with made-up participant numbers, so what else did the authors make up?
b) it's badly done science, with not even basic sanity checking done on the inputs, so what else did they get wrong that the two reviewers and the editor gave a pass on?
Is there a redeeming third option I've overlooked?
Offended and laughing about it.
(Score: 4, Interesting) by looorg on Sunday September 10 2023, @07:51PM
The number of participants is quite low. But then it seems from the article that it's sensor data of neural activity so there could be a lot of data. Otherwise I'm not sure why they would need machine learning to process it. Unless you need it to give the shine of impressive hard computational science. But I'm sure they didn't need it to calculate the gender split, even tho they somehow managed to do just that.
N=33 is a bit low for a participant study, but it might be ok. After all if the patterns are very similar and you notice that by adding more and more people you don't get better data then 33 could be enough. Nothing becomes instantly good or better just cause N is a large value.
That they don't math out to 47% is probably just sloppy rounding errors in multiple levels and someone to stupid to react to it when they read/wrote the paper. Either that or one or more of the participants refuse to identify along the normal binary genderlines and instead insist on being some kind of non-binary-special-entity. Thus screwing the basic elementary school division problem ...
(Score: 1, Troll) by krishnoid on Sunday September 10 2023, @08:27PM
One of them was nonbinary or genderfluid, so maybe they counted as half.
(Score: 2, Funny) by Mojibake Tengu on Sunday September 10 2023, @10:40PM (1 child)
One of the participants is gender unstable, oscillates between male and female state at certain uneven ratio. I am too lazy to calculate the apparent phase exactly.
Rust programming language offends both my Intelligence and my Spirit.
(Score: 0) by Anonymous Coward on Monday September 11 2023, @01:09AM
So it's a quantum thing.
And maybe it's like Fifty Fifty [youtube.com].
Cupid is so dumb.
(Score: 2) by maxwell demon on Monday September 11 2023, @06:12AM
15.5/33 is pretty close to 47%, so maybe one person changed gender in the middle of the study.
The Tao of math: The numbers you can count are not the real numbers.
(Score: 2) by inertnet on Monday September 11 2023, @07:22AM
They recruited 33, one must have left or given invalid results, leaving 32 people. Now that 47% makes sense again.
(Score: 3, Funny) by Entropy on Monday September 11 2023, @04:00PM
Each participant identified as 47% female, 3% male, and 50% other. Remember there's no biological basis for gender, or so
we're told.
(Score: 5, Touché) by Opportunist on Sunday September 10 2023, @07:29PM (1 child)
If it's hyped and pumped by some publisher, it's gonna be a hit.
(Score: 3, Interesting) by JoeMerchant on Sunday September 10 2023, @07:44PM
That was the secret of Harry Potter's runaway story access... Scholastic paid far too much for the initial rights to not promote it to the max, so they did.
Similarly, the movies were expensive to make, so again they were promoted to the max and became the runaway hit of the century.
It wasn't a bad story, either, but there are plenty of better ones that never made it off the slush piles.
🌻🌻🌻 [google.com]
(Score: 2) by SomeGuy on Sunday September 10 2023, @08:08PM (4 children)
So, the next step....
All new songs will be run through this magic AI and shot dead unless they come up as a calculated hit.
Then the calculations will be refined based on which of THOSE songs are a hit.
Lather rinse, repeat, until all we have left are "songs" that have narrowed down on whatever grating hypnotic mind-control bullshit has been shown to be the most successful. And all of those will have been produced by AIs anyway.
(Score: 3, Touché) by krishnoid on Sunday September 10 2023, @08:34PM (3 children)
Then subtract that set from all music, and listen to the remaining set. Problem solved!
(Score: 4, Interesting) by hendrikboom on Monday September 11 2023, @12:56AM (1 child)
Problem solved only if there is anything else still available to listen to.
(Score: 3, Insightful) by krishnoid on Tuesday September 12 2023, @12:14AM
Interesting you should point that out, because there's always people [sxsw.com] who make music because they want to. I remember a (mis)quote somewhere that "The only people who make money in the music industry, are the ones that don't play an instrument." So the music will still be out there, and the Internet will still make it discoverable, but you'll have to listen to it yourself -- or find a way to let your own personal AI select stuff for you to listen to, I guess.
(Score: 2) by Reziac on Monday September 11 2023, @03:04AM
One is terribly tempted to sneak in some different samples, say equal parts Lawrence Welk and Psyclon Nine.
And there is no Alkibiades to come back and save us from ourselves.
(Score: 2) by Gaaark on Sunday September 10 2023, @08:23PM
Can it pick the next President?
And can it be Pee Wee Herman, PLEASE!
ah, shit. he dead.
RIP, Paul.
--- Please remind me if I haven't been civil to you: I'm channeling MDC. I have always been here. ---Gaaark 2.0 --
(Score: 2) by turgid on Sunday September 10 2023, @08:59PM
This can tell me what music to avoid. I already avoid BBC Radio 1 and Radio 2.
I refuse to engage in a battle of wits with an unarmed opponent [wikipedia.org].
(Score: 2, Insightful) by shrewdsheep on Monday September 11 2023, @07:41AM
TLDR; they use ensemble methods on a training sample of just 33 individuals. That cannot possibly work. If people come up with magical accuracies like 97% most likely they did badly overfit. Didn't check their valication step, though.
I wonder what the definition of a hit song is. The underlying population seems to be clustered into different modes of preference (i.e. pop vs. classical). No single song can reach everyone but if the definition includes a song to exceed a certain number of listeners, the definition depends on the composition of the underlying population, making it non-transferable.
(Score: 2) by oumuamua on Monday September 11 2023, @01:35PM (2 children)
What this really means is they can now train music generating AIs without human-in-the-loop. So far humans have had to listen to what AIs made to see if it was good or not - a very slow process has now been sped up. Hope you musicians were worried when AIs started making images because music is next.
(Score: 2) by turgid on Monday September 11 2023, @08:59PM (1 child)
Douglas Adams foresaw the electric monk. Do we need an electric pop music radio listener?
I refuse to engage in a battle of wits with an unarmed opponent [wikipedia.org].
(Score: 2) by maxwell demon on Wednesday September 13 2023, @03:35PM
Well, we already have electric spam readers (known as spam filters) and electric ad consumers (aka click fraud).
The Tao of math: The numbers you can count are not the real numbers.
(Score: 3, Interesting) by ChrisMaple on Monday September 11 2023, @06:50PM
Music that's rarely heard won't sell. I know 2 examples of songs that would never have been hits without large exposure created by unusual conditions.
The Yougblood's first album had spotty success based on the single "Grizzly Bear". Many months later, NYC's WABC radio frequently aired a public service announcement featuring the Youngblood's song "Get Together" from the same album. "Get Together" became a hit and was covered by many other artists.
Then there's the (perhaps apochyphal) story of the Kingston Trio's "The Reverend Mister Black". A Detroit teacher wanted to show his students that they could have an effect on the world. He asked them to vote for the song they disliked the most, then call local radio stations requesting that the "winning" song be played. The result was that this not particularly good song got a lot of airplay, which encouraged purchases first locally and then nationally.
Familiarity breeds.
(Score: 2) by maxwell demon on Wednesday September 13 2023, @04:50AM
I suspect that less than three percent of all songs become hits. If so, I can beat the accuracy of that method with the following algorithm:
The Tao of math: The numbers you can count are not the real numbers.