Comment FAQ: Generative AI defense comments: Arguing that your Suno/Udio music is equal to my music was already done in 2023 when a federal judge decided that AI generated content isn't copyrightable nor qualifies for the same protections as human-made content. If you have a valid argument for this, then go appeal that decision. This is not the place to argue about this. Go read the actual, concise, deciding argument by searching "Case 1:22-cv-01564-BAH". My "ai detector" model use: I've already had some calls with data-certifiers/content-ID companies that I had previous association with and it is completely possible to integrate this AI detection into their APIs. What I would much rather see happen is distributors release a blanket policy stating that they will not be paying royalties on content generated on Suno/Udio/etc and directly uploaded to their distribution, and will remove/chargeback royalties earned from that material.
Direct link to case if anyone else was annoyed by UA-cam not letting you copy and paste on mobile: www.copyright.gov/ai/docs/district-court-decision-affirming-refusal-of-registration.pdf
hi, your points on how much the streaming services pay per stream got me interested: what's the pay on bandcamp like? i'm sure less people buy music there, but a single album purchase must be worth hundreds if not thousands of streams, right?
@@BennJordan "For the foregoing reasons, defendants are correct that the Copyright Office acted properly in denying copyright registration for a work created absent any human involvement." Yes, for THIS specific case, not for AI music in general involving some human involvement. You just cherry picked what you want: the reality about copyrighting music made with AI is different from what you think it is. Again: do you really think Suno terms of use are illegal ? Prove me wrong.
If people enjoy listening to AI and can’t tell the difference between a human artist why shouldn’t they get paid. Someone still has to do the work of generating a “song”.
As a graphic designer AI art has been absolutely unbearable and really affected how i work, i've been forced to use this chrome extension called 'Pre-AI Search' which literally just filters to searches before 2023 (lol) and its the only way i can actually get non-AI reference designs anymore. Can't even imagine what the internet will be like in a few years.
If you're on a different browser or don't want an extension, before:(year) works in Google to show images before a certain date, I.e. (your search) before:2022
My graphic design work went from 60% of my work to less than 10% in 2024, the hit is real. Thankfully as a long-time DJ (since the 90s) i have been able to take on more gigs to cover that (don't get me started on the state DJing is in the last few years hahaa).
You can detect AI-generated artwork basically the same way. Even when you download AI-generated art as what is supposed to be a lossless PNG file, it still has JPEG compression artifacts in it, because the data it's been trained on is virtually all from compressed JPEG images, so that's what it thinks the images are supposed to look like, and thus the AI model reproduces the compression artifacts as part of its own art.
@@agentbananauk if a file is converted from PNG to JPEG, and then back to PNG, it will have JPEG artifacts. Most JPEG artifacts aren't very noticeable, so it's unlikely to cause problems, but it means you can't just say "this PNG file has JPEG artifacts, it's AI!!!!!", because it could've been converted from JPEG to PNG along the way. The only way for this advice to work is if you get the file directly from the "artist" and it has those artifacts, or if the "artist" uploaded the image to a site that doesn't convert to JPEG (Twitter/X does convert, so not there)
Most of the AI detection tools that try to do anything remotely like this have abyssal accuracy rates, and those success rates are generally higher with older, outdated generative models. That includes both false positives as well as false negatives. And the moment any editing is done to the image (cropping, resizing, transforms, filters, upscale, etc) their accuracy drops even lower.
My biggest issue with users of GenAI is hypocrisy. I know programmers who will happily use Suno/Udio while simultaneously railing against coding assistants like Cursor which are trained on tons of copyrighted data. I know musicians who happily use Midjourney/Stable Diffusion to generate images while simultaneously cursing Suno/Udio. You don't get to cherrypick which form of GenAI is morally acceptable. Either they *all* are, or none of them are.
Agreed, 'I am' mediocre in so many things I could not do before. I can scrape a website with python or make some generic image with stable diffusion. Im not really mad at it. The sad part is how people respond. It's like being mad about some cookie cutter pop artist. It's sad that people love that, not the thing itself. Udio/suno are neat but I have not heard anything that I would let exist as it does. I mean, it lacks that original feeling. That said, if your workflow was to generate something there and then chop it up, that's kinda cool. Then we come to where I actually get excited about this stuff. Stem separation! Freaking magical! The best thing about AI for me. Then, go on HF, and there are like 40000 niche things and Image detection/identification, background removal, then really crazy specific things
To me it depends on the usage. I don't think it's unmoral for a musician to use genAI to generate a cover image for an album, if the music was recorded by them. The product is the music they made, not the cover image.
If AI generations were respectable you wouldn't see so many people trying to launder them. Everyone knows they have to hide the fact that their so-called "work" is AI.
I think it is a tool that can be used in combination with human creativity and knowledge, it doesnt have to entirely replace it or be devoid of human soul or effort. I write all my own lyrics and draw all associated art for my music- album cover, profile pics/ banners etc with no AI assistance. I use AI mostly to generate vocals which i can control the pronunciation and flow of. All I want is some appreciation for my lyrics and AI allows me to present them as a fully finished song which I have a ton of control over. its the best medium I currently have access to be able to share my music right now, so I am grateful for that, and i am proud of what I made. I cannot perform vocals where I currently live due to shitty circumstances and environmental limitations. I use AI and try to retain as much authenticity as i can to share my work, and i think i accomplish that skillfully, better than any other AI music ive heard thusfar.
@@ITZTLI-music Dude, then make instrumental music. Like I did when I couldn't sing. If the issue is only technical and not that you actually cannot sing there are many many ways to solve with without resorting to so called AI
@@ITZTLI-music so you're supposed to find someone to do the vocals, not find an app that is using someone else's vocals without their knowledge or compensation
This stuff makes me super sad. All of a sudden on Spotify my related artists went from a bunch of artists in my general space to being all AI artists and it just makes me feel like shit. Discovery was already horrible on Spotify and now it's just impossible.
I've always used UA-cam for music because my taste is someone niche, but even it is getting overrun with AI music. Trying to find a playlist of genuine, non-AI classical music is way more difficult than it has any right to be. And "classical" AI pisses me off even more than others, just like the ones that rip off renaissance paintings and try to pass them off as genuine. As if grammar and spelling aren't already at an all-time low with the internet; our knowledge of art & history is damn near done for if this shit isn't fixed.
The low pay per stream combined with the growing low quality content has moved me towards alternatives. I don't stream anymore and instead support my favorite artists on Patreon and I buy music from the artist (often on medium like disk or tape) on Bandcamp. I feel better knowing the artist at least earns something and I much more enjoy the music I listen to. Remember that time you went to the music store to buy a CD for that one song and you kinda start to appreciate the rest of the album too? Well: that feeling. Please keep making these videos, they are educational and a joy to watch. Thanks.
I always buy FLAC (if they have it) or 320KBPS MP3s of what I like, and I download them right away and put them on a thumb drive. I don't feel like I own what I bought until I can listen to it without a website giving permission.
@@FourWaveforms"I always buy FLAC (if they have it) or 320KBPS MP3s of what I like, and I download them right away and put them on a thumb drive" - For anyone who doesn't follow this advice, definitely start doing so, especially on Bandcamp. Do NOT assume that because Bandcamp has an app with which you can listen to your library, that it will stay there forever. Artists can, and do, remove their music from Bandcamp (or remove individual albums, or individual tracks from albums), and when they do, it's removed from your library. If you didn't download it when you had the chance, then it's gone forever.
That certainly a good alternative. However, I think artists earning money from art will be going away. Being able to scale earning from your art quite modern, and I honestly think that making art for art's sake (instead of for likes or money), is the core of art, and I don't think art it will go away even if nobody listens or pays. Bach or Kafka would still make their works.
@TwilightThrone but Bach was a church musician who made money from performing. I dont think if he was preoccupied with other work that he could have been as productive as he was. The way I know that is that almost none of the most famous composers in the history of the western musical canon worked with something other than music as thier career. It will be a massive loss when no person can survive on thier art alone.
Using the concept of missing song data from file compression to catch AI is FRIGGIN GENIUS ! You're officially the "John Connor" in this geeked out robot battle. 😊
I'd love to have a browser plugin or something to get to use that AI for myself. I've seen SO MANY AI songs that aren't caught in the playlist linked below the video (because there are probably hundreds uploaded each day?).
Also, I didn't understand how the "fingerprint" was created to find AI music with 100% certainty; even though I somewhat know how LLMs and other types of AI are created.
i saw a similar thing of catching ai generated images. ais incorporate so many compressed images into their training data, they mimic a lot of compression artifacts in their outputs that arent natural or dont make sense when they mash together different compressed images in the training set. because these compressed files are so baked into the training set, it doesn't seem like it is possible to remove these artifacts from ai models unless they were to retrain their models on only uncompressed files.
Longtime Qobuz user here. Glad to hear they pay a little better (though as you say they don't have the volume as they're a niche player). But the other main reason I like them, aside from the excellent sound quality, is that you can purchase the music, either tracks or whole albums, and get digital downloads (FLAC, mp3, etc.), DRM-free, to keep. And I imagine those pay artists better than the streaming. So I regularly buy music on there that I'm finding myself listening to so I know it won't disappear, and hopefully compensating the artist.
Yeah also a big qobuz fan here, I've found some music on their site that I just couldn't find anywhere else and bought it, paying the artists big bux and getting some high quality shit. I fw it🔥
What helped me choose Qobuz briefly, was that Qobuz pay per view being the highest. Tidal lets you give more money if you’re on their highest plan to your most watched artists.
not to mention that, unlike Tidal, they have FLAC for most of their catalogue. (Tidal uses MQA for much of its catalogue, has been called out on this, but never did anything about it, making them commit false advertizing)
i joined an ai music group to spy on people and yeah, these guys are posting their music on spotify and trying to claim themselves as artists and music producers. some even sell their ai songs as legitimate beats to real people. some have gotten in trouble and their ai songs have been taken off spotify, but they legitimately see themselves as artists. ive tried to listen to a few but they all sound like trash. left the group because i couldnt stand the entitlement of these people.
@@KwissBeats Did you watch the video? AI created music is not copyrightable in the USA. People are taking unaltered AI generated music and committing fraud by uploading it as their own creative work to establish a copyright claim and receive revenue. Artists that do this are not "artists" and are lying to the public. Separately, calling music generated with loops you own or possess a license to is not fraud, is copyrightable (for your work but you may not exclusively claim copyright to the unaltered loop) and Benn points that out as the "Splice Problem" Venus Theory originally named. Benn clearly says the problem with loops is the content ID system with media platforms attaches copyright exclusive ownership of the loop (which is not correct) to the first user on their platform with the registration and neither the loop publishers or media platforms can be bothered to implement the simple, but not free solutions he illustrates.
As someone making ai for a living the past 6 years, 100% accuracy is at best an indicator of something gone way wrong in methodology. I like your content Benn but be sure you are not advertising something over confidently! I know I have in the past. Data leakages and not using good test/val datasets happen to all of us
That being said, if there is actually a signature in ai audio that is 100% distinct from the rest, then I really admire that you found it! Maybe cnn are overpowered for the task too :)
@@mrolowu1 so that's what it is about... I tried it and it just started adding random junk I would never listen to. I dont like lyrics - it gives me lyrics. Why?
@@koppy82 it's literally already been proven, I'm not saying what you said isn't happening but they absolutely are publishing ai music on their platform themselves
I appreciate all the people on a music channel straight up coming out and stating they have absolutely no love for music, art as a whole, or the artists who create it.
I haven't finished the video so I apologize if this is mentioned, but one thing the GenAI boom has also facilitated is the age-old tradition of buying listens/likes/views/etc. A while back a friend - who is in a fairly successful band - and I were talking about another band from our scene that got signed to a major label before ever playing their first show. Now, obviously they had connections, but then the label has to make it look legit. My friend was like "oh yeah, here's what they do," then pulled out his phone and pulled up the band's Spotify page - all of their suggested "similar artists" that fans "also listen to" were AI-generated "bands" from Eastern Europe that only had one song each, each with tens of millions of "streams." He explained that you can just rig the whole thing with AI, buying millions of streams without anyone ever having to actually listen to anything, but there will always be signs when an artist does that.
Capitalists just filling the world with garbage to make a quick buck. And odds are that band will just be chewed up and spit out and it’s on to the next one. Fucking vampires. I can’t see being the artist to agree to that deal, but I really can’t imagine being the guy at the record label in charge of creating fake credibility for the new act that just got signed. You can’t take enough showers to wash that off.
Your content is a (good) outlier in the world of synth/music UA-camrs. Even if you think some of your material is too nerdy, complex, or weird for standard audiences, I encourage you to follow what you love.
It is a little misleading to say that compression throws out, say, 80% of the INFORMATION. It throws out 80% of the DATA. In theory you can compress a one-hour recording of a sine wave using two numbers, the frequency and the duration: e.g. sin@440/3600. That is WAY less information than a 128kbps hour long recording, but the reproduction would be BETTER than a transmitted recording. All sound can be reproduced as sine wave combinations. So, while clearly music requires more than one formula, compression does not PER SE mean information loss, just data reduction. Further, sampling error forgiveness is not JUST a psychological effect. For example, so long as the sampling rate is twice the highest frequency, Nyquist and Shannon showed that there is a mathematical stability to a sampled result--a result that can lead to compression mechanisms that do lose information, but not at the same rate as they lose data, for example by first filtering out barely audible high frequency content.
Great clarification about data vs. information. To say a bit more about the Nyquist-Shannon sampling theorem: while it guarantees that frequencies below half the sampling rate can be faithfully reconstructed *in theory* -- the practical quality of frequency reproduction depends on factors like bit depth, filter design, and other implementation details.
But that's not how Mp3 or any similar lossy compression system works. It absolutely throws out information. Sure, it isn't necessarily information you would have heard so you could say, in a sense, that is was redundant, but that doesn't mean it wasn't information. If it was a graph of share price, for example, rather than music, then you wouldn't say that throwing away 80% of the data did not lose a large amount of information. If the information was still there, you wouldn't be able to do a delta between lossless and lossy compression.
Lossy compression clearly throws out information. Just encode anything on the higher frequencies of the audible range, compress, decompress, decode back to regular frequencies, and you'll see how much worse it is.
Great explanation. Another way to think about this is JPG vs SVG. A rasterized version of a vector image might be 5MB, but it's vector original only 50KB. The vector image represents the same contents in different way, leading to similar, or even better reproduction at higher zoom levels.
I mean... You can stream from it x) But only the albums you bought to begin with. Similarly true for dizzylab, which is a Chinese "replacement" for Bandcamp since BC got banned a couple years ago.
Probably no, unfortunately. It's technically one of the worst as you can listen to an artist for hours without the artist earning any money from it. I personally don't use any paying steaming service and listen to my music primarily on Bandcamp and on UA-cam. I've bought some music on Bandcamp, but just a tiny fraction of what I've listened to more than once because 5-7€+ is pretty engaging. It would be interesting to know, for The Flashbulb for instance, the approximate pay per stream if we compare the number of listens of 2024 to the total of purchases over the year. I have really no idea how many people on Bandcamp actually buy their music and if that model is working well for the artists. As you say, it's a very different model, Bandcamp being primarily a retail store which you can use as a streaming service too. But perhaps it shouldn't stay that way. I would personally gladly pay 10€ a month to Bandcamp to have say 9€ being distributed to all artists I've listened to during that month. They are perhaps afraid of going in that direction, idk.
hey Benn going off the happy note of the end of the video, I did want to say, I have recently canceled my Spotify subscription and have no intention of going back. And your videos on Spotify and how they treat their artists were the main reason I did so. In its absence I've been buying most of my music off iTunes which might not be the best as a large portion if not all of the money may just be going to the labels, I am still happy to have made the switch off Spotify and will keep trying to find ways to more directly support the artists I love. thanks !
Disappointed in many of the comments on this video. Where is the beauty in buying creativity from a company of thousands of engineers rather than cultivating it on our own? No one can convince me that I'd be better off with this garbage product than with my own hands. I don't care if I don't produce anything worthwhile, at least it's me producing it and not some amalgamation of stolen material.
Think of it this way. If you, like me, have shared a talent (musical or not) with another person, you might have received a comment like, "That's so cool! I wish I could do that/I always sucked at that/I'm not smart enough to do that/etc." These people usually genuinely believe these limitations are something we were able to surpass via a stroke of fate or a blessing of some sort, or that these limitations are simply given only to some people and are not surmountable. Alternatively they're self aware enough to know that with practice they could probably get to our skill level, too. And the biggest thing to remember is that that desire is legitimate and often even stronger than you'd first think. But many obstacles stand in the way and they're many of the same ones we all start with when learning a new skill: fear, lack of energy, lack of willpower/ability to care enough, low self-confidence, sacrificing time, spending money, and others for certain. But what if you could hack off a lot of those obstacles? Well, what if the odds of winning the lottery were suddenly raised by 10,000%? A lot more people would win the jackpot. From there, it becomes a matter of values and ethics. Artists value artists. Not everybody, though, values artists, or at least values them enough to have those strong ethics that give the person the desire or willpower to refrain from using things like generative AI, pirating independent artists' music and small artists' art/films/writing/games/etc. These tools are truly appealing to many people. So what gives a person reason to choose NOT to use them, let alone advocate against them? As a society we don't generally respect artists the way we should. This is evident enough in the art/media industries/economies.
Lossless and Compressed are not mutually exclusive. It would be more correct to use the terms lossy and lossless to differentiate. FLAC (lossless compressed audio codec) is identical to WAV PCM (lossless uncompressed audio format).
And almost all people wouldn't be able to tell the difference between that and it run through a well built compression algorithm. That's the point of "Lossless"
@@brianbauer3148 the only way I can tell an MP3s isn't compressed well is if the cymbals/hi-hat sound wrong. But I have some 192kbps MP3s that sound just fine
@@brianbauer3148 why is lossless in quotation marks, it's literally lossless, it's lossy compression that's built on the idea that u can't tell the difference anyway
there is also the possibility that the experiment they mentioned had flaws. we dont know if the equipment and song choice and people had any affect on the sounds. plus the song itself could be lossless but still have lossy samples . and some people like myself whom are sensitive to sounds can really tell when audio sounds off. i for one cant stand lossy audio. at least when its compressed past a certain point. the diffrence is there but i think its not "most people dont notice" likely its "most people havent experienced the difference and are used to compressed audio as the norm" i mean most of the audio consumed nowadays is trough compressed social media sources
The Splice problem is not just a sample site problem, it can be a synth sound patch problem as well. A friend made a track using a preset on Reaktor. When I heard it , I asked him why did he decide to sample a certain artist. He had no idea what I was talking about. He then showed me the preset he used. I then played the track (a pretty popular track) and we both were floored. UA-cam would’ve definitely flagged him even though it’s a preset sound and not a sample.
Starsky Carr has several videos addressing this where he attempts to demonstrate filter sweeps on analog synths and they get copyright struck by someone who has taken Moogs, Oberheims, Prophets, etc. and just swept the filter slowly. That kind of BS can definitely threaten UA-cam as a platform for gear reviews. UA-cam gear reviews have plenty of problems (lack of transparency of financial support to reviewers for one) but taking a raw saw wave and sweeping a low pass filter ought to be an easy "non-copyrightable" sound. Sweeping a filter over a raw wave doesn't meet anyone's definition (except UA-cam) of a "transformation."
@@scott21113 META is adding ai and bots to their platform. AI has been slowly increasing in frequency in my feed every couple of months across ALL platforms. If the system doesn't collapse under itself or the AI becomes self aware and destroys their own makers a la Skynet, I don't foresee anything happening. The big record labels don't have the balls to take on Google, Meta, and ETC who are accelerating the rot. First Mistress Orange Trump assumes office soon with Lord Elon wagging his money collar AROUND THE WORLD putting his short nubby fingers in every sovereign nations "honey pot." Have you noticed so many platforms have began sucking off these psychopaths the last couple of years trying to get in the good graces of the hopeful plants they paid for on the campaign trail? AI has and will continue to be a massive mistake, but the pandoras box is open. The west will fall before it regulates itself to ensure it doesn't self immolate, its practically THE American pastime of voting against our own interests. I would be shocked if the shareholder suits banded together to right the ship they've set to sink slowly over the last decade.
It should be noted that streaming revenue differs vastly by country and subscription status. Spotify free from Argentina pays 30x less than a paying subscriber from Switzerland. Also, UA-cam Red from Singapore pays very well, about the same amount as Amazon Music in USA. My dataset is pretty small, but overall it checks out with what Benn said. Spotify genereally pays about 0.0018 from countries other than the 10 richest western countries. I failed to install any Chinese streaming apps, but some of the music that got submitted to them just disappeared. (Again: I'm a no name brand musician that made a few ambient songs so probably just got flagged as spammy meditation music and blocked by their automated process, and not because of some evil scheme to steal my half a cent per year.)
Funnily enough, the sole exception is russian VK. They pay pennies to their own content creators, music, video - all the same. Just a pointless platform
it also differs by subscriber type. pelaton users only listen to their streaming platform while on their bikes. this severely limits how many total streams users have. divide the already low streams by the high monthly cost and you get a very expected significantly higher than normal pay per stream. the metric is just dumb. it’s used by people who want clicks on their video/article/whatever
Your videos are extremely relaxing and thats probably my favorite thing about them. I think its because you passionately speak about what youre talking about, but also that you keep it entertaining and smart enough to invoke interest and learn from. Its like an ASMR video that helps me learn about music production
Such an elegant solution for identifying if something was created artificially. Just getting into your videos. Not a producer, but programmer. This scratched my brain in a new, awesome way
look up their history and realize that the only reason they exist is to shield a few from competition.. they are constructs to to allow a few to maximize profits without the market being able to correct this, as competition is prevented from doing the same thing. That is the definition of a monopol.
@@joansparky4439 Patents were designed to promote R&D, so others wouldn't copy your 5 years of research the day you make them public through a product. Similar for copyright, you don't wanna write your masterpiece and have hundreds playing it without you seeing a dime. That said, these reasonable ideas got lobbied into a monster that keeps corporations profiting off dead men for decades after they're gone, while holding culture and technology back from running with these new ideas for entire generations.
This video is what’s up! Another thoughtful and relevant piece. You’ve convinced me to both sign up for your Patreon and, push to to inspiration from your work to make more meaningful content on my own channel. Long time fan Ben, keep it up.
Just discovered your UA-cam channel today, this is my second video and I realized oh shit, you’re flashbulb! Been listening to your work for a while now
Love it that Benn and Venus Theory had a talk one day and they both decided to start giving zero fucks and being way more honest and freewheeling in their videos. Love it.
@@BennJordan Are you willing to share any details about model/method, training dataset size and number of epochs, validation set size, etc? A 100% accuracy result on a classification model is generally a very bad sign, and what I could find from the academic community (mostly audio deepfake detection, not necessarily music) was in the 90-95% range.
@@spacepxl Yeah, for classification models 90-95% is good, 96-98% would be brilliant, 99-100% is at least concerning. I agree with you: it would be nice if they share with us the pipeline in the future to replicate the model using our own training data
@spacepxl honestly, I thought even putting that little fingerprint image on the screen was too revealing. I hope he keeps it secret for as long as possible lol
@@BennJordan I think you have to keep in mind that my father's generation was also raised on music and were very involved with Hi-Fi systems. My father to this day isntantly recognizes Mono sound even though I think I have the better ears. But they were raised on these media similar to newer generations being able to tell AI voices apart from real ones much better it's a matter of how early you got into contact with it and how you trained your ears. Unfortunately many people do not consciously listen to music anymore. It has become just another commodity. But there is one huge difference with compressed files and lossless. Moving pants in front of the subwoofer a DJ told me. He always can tell if it's compressed when the air isn't really moving due to the lack of infrasound. And I would bet that if you showed younger people comparisons they would get better over time. That is not to say that they would be able to tell the difference with the higher bit rate. But they surely would be a whole lot better. I surely did.
Interesting about compressed vs lossless audio. I general I have noticed a difference between FLACs from CD audio vs mp3s. Usually the cymbals get distorted first and you start to hear a watery noise as the quality drops, as if you have water stuck in your ear or you're listening from underwater. I've never done a blind study though but I'd be interested, but the difference is there to hear. Unless you've spent time comparing A and B however, you may lack experience to make that judgment. I also suspect there are times where the right bitrate of an mp3 and the right frequencies used make it less susceptible to being analyze in this fashion by a person, I suspect.
It’s like wine tasting, just because most people drink wine and think it all just tastes like grapes doesn’t mean there aren’t flavor profiles one can easily distinguish if they know what to pay attention to. Similarly music listening is a passive activity so most people listening to lossy audio streams just hear “the song” they’ve always heard and liked… it doesn’t mean there aren’t musical layers in the signal being thrown out that a discerning ear can listen for a hear. I can always hear a difference between lossy UA-cam/spotify music and my Qobuz stream and flac files. To be fair, cheap Bluetooth earbuds that a lot of people use are poor equipment to detect these differences. But us music connoisseurs who have resolving enough gear can always hear it. And the kicker is the higher bitrate streams/flac files always sound more emotional/musically engaging to me. I feel closer to the music
I did a non-scientific non-blind test on myself compressing down a ripped cd track (can't remember the genre but it was a good mix and production, before the loudness wars) at 320, 256, 192, 128, and 96 CBR stereo mp3s listening through low-mid range headphones. I recall starting to hear differences between 192 and 128kbps (edited) as the stereo image seemed to narrow. As you say, cymbals seem very susceptable to distortion in lower quality mp3s, and I hate that sound
Regarding the last statement. I left spotify in 2020 and within the year had bought most of my favorite albums I found on it via Bandcamp Fridays. To this day I still keep an eye out for it to come around and happily pick up an album or two for the money saved from not paying for a streaming service. I couldn't recommend it more to anyone else. And it's mine. It's there when I want. I get to enjoy it in all the FLAC glory that my ears are too stupid to hear.
I still buy physical albums in the form of CDs when I have the option to, I rarely actually listen to the CDs but I rip all of them using the FLAC codec. All my bought digital music comes straight from Bandcamp. I just like the idea of owning my music, and giving the most amount of money that I can to the artist. All the physical CDs I have go back years so they also hold a big sentimental value for me.
Bandcamp is the best. You can choose from a variety of file formats. Your music is always there, if you need to re-download it or choose a different file type. Once you buy it, it is yours. No DRM. The meta data is usually flawless, so it shows up correctly on music players, and like you said, it feels great to buy music on Bandcamp Fridays. I can't say enough good things about them. With that said, it's so frustrating when artists don't put their music on bandcamp, because then I'm forced to go scouring the internet for how to buy a digital album from somewhere other than Apple and Amazon, which is bafflingly hard. I will only buy music from amazon as an absolute last resort. This led to my discovery of 7digital (kind of terrible website and download process) and highresaudio, both of which have limited music libraries. I have had luck finding albums on Qobuz that weren't on bandcamp, as well, which I feel better about seeing that they are the top-paying streaming music site. I might go the route of buying cds again, but PCs don't even have cd players/burners anymore, which means that I have to buy something external. Streaming has made buying whole digital albums so convoluted. Don't even get me started on digital music players and all the annoying UI flaws that they all have. When streaming became a thing, it's like everyone just collectively gave up on continuing development of the hard drive digital music experience.
@@pronstorestiffi Same same same. If not raiding the localish Record shop (Wuxtry) a couple times a year I try to order directly from artists on BC. The nice thing about buying local is that most of the time you get an included BC code card as well these days which is really nice. And regarding the "joke" about lossless and compressed, I'm aware and can tell the difference between MP3 compressed and WAV/what have you.. Hell, it drives me insane when making music how I can hear the compression happen from render lol. It's just that at a certain point it gets silly to me personally as I can't tell the difference. Even after trying some *very* expensive sets for funnies in testing. My dailies aren't the best being a set of DT990 Pros driven at 250ohm properly. But they're enough to make it obvious I'm listening to UA-cam VS local 99% of the time.
Please create a music streaming platform that is fair and just and I feel that you are the man for this! I would help with whatever I can (UI/UX design, graphic design) to make this happen and I mean it. I think that others would chip in too, it could be a community project for the community. Thanks for all your great work!
Most (or just about all) of us complain about AI in the artist realm from the comfort of our lounge couches. And then there are genius heroes like Benn and Cameron who does the work. I wish I had half of your brain and all your guts Benn. Your videos are SERIOUS game changers - keep the light burning.
That's not an expletive, that's a body part unmentionable in polite Victorian society! Nice WTF moment, makes you wonder exactly what goes on in Benn's. Patreon content...
You know, . . . I just grab a musical instrument, any instrument, and just start playing. The business of business and the highly-processed product consumption . . . just makes me ill. Hang out with friends, man. Tune in to the stories and rhythms that are happening with each other right now. It's so much richer to be alive than yanked along by faraway sound-hook factories.
I really like this. But also some people fell in love with the recording arts. Hi Fi gear. Preamps. Tweaking knobs. Finding mic positions. Lost in time like tears in the rain...
Not all of us can be an extroverted “people person” who wants to collaborate with other humans. I play real guitars and I’ve played in bands. I currently spend most of my free time designing and building my own custom instruments out of wood and metal. Mostly with my hands. I’ll get something cut by laser or CNC if it makes sense and I’ll use samples generated by AI if it makes sense. I get a lot of joy from the things I create on my own, on my computer while not having to depend on other people or compromise my art by letting others determine what it should do or whether it’s good enough. Nobody gets to tell me that my music is any less valid than the next guy’s. If you don’t like it, don’t listen to it.
Oof. Hard disagree. The amount of enjoyment i get from producing art depends heavily on whether or not others are enjoying it. If im not inspiring somebody or bringing somebody happiness with my art, then im failing as an artist. Mind you, there’s nothing wrong with your motive for art either! Im just so tired of listening to this stale ArT mAn take that insists art has to be some kind of masturbatory practice in order for it to be authentic or real.
In a world where “ that was inspiring” is something I think less than I could have ever possibly imagined, I’ve just typed it in a UA-cam comment on a video that inspired me and meant it. Well done. Well done, indeed.
Hey man love your content. I think you should take all of your videos from the past few years and boil them down into one large video explaining all of the information involved with the music streaming scam industry. I dunno. Anyway keep it up!
Benn your channel is so awesome. I am not a "music enthusiast" nor a musician, but I really enjoy the science behind your videos. You do your research and bring a genuine, honest air to whatever subject you're approaching.
Having used Suno for a fair amount of songs, about 500~ or so, I've noticed that in almost every creation there is a weird residual sound that appears in its creations. It sounds like a bleeping synth and it usually occurs in parts where the song is becoming a bit more calm, so right after a chorus or before a buildup. If you hear that sound, you can immediately know that it's AI generated.
instagram already tried integrating an AI detection system and it caused a bunch of people to think artists they loved were using AI when they weren't because AI gets lots of stuff wrong constantly which people will know from how often Google's AI confidently gives a wrong answer
The "detection system" was literally just reading meta tags from edited files and those tags are only placed there if you use a generative AI feature before saving the file. In literally every case I've seen it's been admitted to eventually, usually with the user saying they "didn't know" that one feature counted as AI. There was no detection algorithm, literally just reading the metadata to see if the file had been tagged. Photoshop only tagged files that were in-fact modified using Gen-AI features. So no, nobody was wrongly accused via that system and anybody who says they were is about as believable as your average video-game cheat who says the same thing. "It wasn't me I swear! The system messed up!" Sure thing.
Found your channel and bought the zoia because of you! Your interests and experiments etc are very inspiring thanks Benn... you're adding massive value out here
I think the work you are doing exposing serious problems in the music copyright system is very important. Hopefully, this will be dealt with fairly, eventually. Content ID has been abused already by bad actors that steal monetization with no downside risk to themselves. Now, with AI music production in mass, it is spinning into a fraud race to the bottom that threatens all creativity.
How has youtube never recommended me this channel before. I already watch stuff like this but its just more and honestly a higher quality than most channel.
okay, ill join your patronite, my first patronite ever keep up with the good work, i can see that its genuiently brings you joy, props for being independent!!
I couldn’t tell the difference between compressed and lossless on individual tracks but over time my ears tire from compressed audio. When MP3s came out I loved them until I started to lose intrerest in music. Vinyl and lossless brought my interest back and I can listen all day at home now. In the car on mp3 im good for an hour then it’s podcasts.
@@sinane.y Care to elaborate? Is it too dead of a space? I can assume speaker placement has to do with it as well, as most are in the door down at foot height and some tweeters far in front, close at the windshield.
I only discovered this channel a month ago, so I appreciate I'm massively late to the party but I don't think its hyperbolic to say I think its some of the best content on this platform for my personal interests. Thoughtful, considered and articulate content from someone who seems like a really top bloke. I'm bingeing the back catalog like a boxset rn.
I decided to buy all the music I actively listen to a while ago. It feels better to know that I’m supporting the artists I enjoy, and to own the music I enjoy. Absolutely loved this video, thanks for doing the research!
I don't have a single music streaming service. I tried free Spotify for a week and went back to my MP3s. Thank god for being such a fan of 90-00s Warp Records and 80s heavy metal. It all fits in my Sony player which has a 3.5mm headphone jack for my 12 dollar headphones.
I love me an mp3. Even if its crunchy, or glitches from years of data loss being moved around from ipod, to sd card, to USB stick! Its like that "dirtyness" of an old mp3 tells you a history of it. I think high quality wavs and flacs are great for certain songs, too. but i'm more of just a fanatic music hoarder, instead of a full blown audiophile, and those lossless files tend to be huge. i know i sound like a quanity over quality person, but thats just how i love having my music!!! Love the video man, puts an interesting perspective on streaming services as an artist! Great work!
The other problem with streaming is that everything is focused on single tracks. This means that the depth of an artist does not have a chance to emerge. I still listen to my old vinyls and CDs with pleasure.
This is extremely useful work with lots of applications for a pretty niche field of hobbyists and artists compared to the numbers and views you could pull from chasing a more viral topic with your interests. Thank you for shining a light on how AI has been shaping the monetization of music for artists and people who actually care about the work that goes into music. Even though I only produce music as a hobby its so annoying how people who have a true passion for creating new music can have their work undone by some kid with a keyboard and a subscription fee, built off the backs of thousands of years of artists with visions for their art rather than a chat input.
I would rather get that info from someone who actually makes bots, not some random video maker who has never made a bot once in their lives and likely has no idea what TCP even is.
The ability to detect MP3 depends on the high frequency content of the track. MP3 (and others) break down when there is a lot of high frequency detail, since I listen to a lot of live real drum music, if the drummer rides the open hihat or the ride cymbal it creates a flanger or phaser effect in the MP3. Once heard, can't unheard it. A track with 808 drums can be indistinguishable as MP3 though.
The best way to support musicians is to buy their music on bandcamp or mirlo and then sync it to your music player like it's 2004, or run Plex on your computer and plexamp on your phone. While that doesn't pay per stream, the 80 cents or so that you pay the musician up front goes a LOT further.
I use Jellyfin (lossless streaming my entire archive) and Syncthing (offline listening with compression, think of it like a podcatcher)*. I tried to explain to my quite tech savvy friend just a few days ago that my music archive of mostly *lossless* music is expected to not grow beyond 2TB. I'm 40 now, a FLAC album is about 400 MB, if I buy a new album per week (which I don't), 50 in a year for simplicity, until I'm 60, well that's ridiculously little data. Estimates beyond 20 years don't seem to make much sense. But it's safe to say that storing audio is a solved problem today and storage density will still grow. The compressed version of my archive fits on a finger nail (an inexpensive SD card, the entire library can be compressed from FLAC to Opus in about 70 minutes btw ) today. So I can have everything I *like* always with me. Among that some artists and albums that are not on streaming services for various reasons. I tried to convince him with statistics and data. He didn't get it said "there are 50 million songs and albums (?) on my service". Well I know about GenAI for music, he does not, and I have seen on vacation, when we traveled abroad, that his service contains uploads from other users. Lossy and piracy, and bad tagging if you so will. He does not know what his service has to offer. The amount of inconsistent (user provided) garbage and generated garbage is growing. However telling someone that the service he pays for feeds him mostly garbage while the actual artists benefit the least from a monthly subscription, well that's not winning an argument. Look, I was a pirate when I was young. I knew that what I did was not legal. I developed a sense for quality and I can now pay for things. Meanwhile it looks like businesses are pretty comfortable with piracy as long as it benefits them and they have their PR mouthpieces touting that it's not piracy and a fair opportunity for everyone. *I can add an entry to a text file to "publish" something and the album will be downloaded for me (while still having the option to stream with Jellyfin). Yes this even works from my phone. I'm sure that with today's possibilities of coding something the process could be streamlined a lot more. Plex and Jellyfin don't seem to be much more than a web UI for FFmpeg. It's just that people are lazy and misguided.
@@braintrust12 There are plenty of ways of getting exposed to new music via Bandcamp, and there's nothing nightmarish about downloading your music and keeping it on your own device despite what Spotify would have you think.
Timberland is currently working with Suno so I guess it will get a lot harder to detect it… also I know artists have been using this technology prior to this popular technology. Also you look like Kevin McAllister… I like that, just subscribed❤
Bandcamp?.....does not pay for streaming, is how that sentence ends. If you purchase an album you can stream it, but there is no 'streaming service' on bandcamp and artists are not paid money for streams, only for the purchase of the album. Also, bandcamp got bought out and is owned by Epic Megagames which is killing PC gaming with their predatory business tactics. I took all my music off bandcamp when they got bought.
@ihaveagoddamnplanarthur It kinda is, but not in the same way. I was waiting for some kind of data on BC as well, as I pretty much exclusively get my music from there.. I buy digital albums most of the time and sometimes some merch, like a shirt. It would still be interesting to understand, if and how much artists get when I listen to their material, without buying anything.
Their streaming model is slightly different than whats considered a streaming platform so i guess it would be difficult to compare with others, but i do think Bandcamp and Soundclound are good choices everyone should consider on their de-spotifying journey.
it's seldom the video itself that's the cool part. what it's documenting is what counts, and that still takes human creativity and discernment. Generative AI is by-nature derivative, so rarely adds anything of real value. AI tools are really handy for filling-in the blanks though, but any more than that usually ends with a mediocre result.
@through-faith-alone it's derivative of your work, and of the training. Sure, there can be some pretty interesting emergent results, but that's not the same as creativity.
@@theafro fair assessment but no, it is the same as creativity. What it isn't the same as, is the same amount of creative work, or creativity. Is it less creativity? Sure. Most music is derivative of other music. There's less creativity too. The difference is in the level of engagement, and use of the tool. In the future, you will see musicians using AI with impunity, as many are already. To create outstanding pieces of music. And you will also see people expressing themselves through AI making music for them. Lastly you will also see mass produced music through AI. But creativity will never be lost, it's not the tool, it's the user.
Hey Benn, great video. It would be interesting to see, in your investigation of copyright strikes, if included, there was also a song that was %100 original recorded with live musicians, and see how many strikes that gets. I bet it would not be zero.
The reason most people chose the lower quality music as being the higher quality is most likely due to the fact that the lower quality is their "norm". So many people stream low bitrate music on their day to day, and rarely (if ever) listen to CDs or vinyls, so their mind has made that particular quality the "standard"; and since everyone thinks their way is the best, their mind has set that quality as the best ... You then ask them to compare what their mind has deemed the best to a higher quality, and they're of course going to say that their normal (lower quality) choice is better. The mind is pretty silly. And yes, I dork out on neuroscience podcasts all day at work.
The other side of the coin is the systems outputting the music. Air Poops are all the rage and they sound horrible, meanwhile my Sony, Koss, AudioTechnica, and other headphones completely blow those things into the water, with the competition being 4-5x more with the EarBra Max Pros barely keeping paces with headphones that are DECADES old at this point. That also is part of the problem. The smartphone market only had LG, MoonDrop, and Sony the last few years providing the mobile experience that is feasibly achievable outside of buying fiiO or some other extremely niche device for an already less mainstream niche that is HQ audio....... People these days are spoiled with 10s of thousands options, all of them poorly made crap from china with no prior reference to "good" sound. You won't find the majority of people running Focal in their cars or having 5-7.1 surround for their TV unless it comes in a complete "sound system" box. The thing that gave me hope for humanity regarding music was that Vinyl was making a resurgence before covid and platforms such as bandcamp are readily available these days.
@WhatWillYouFind air pods and old school earphones of the same design are the absolute worst. Horrible quality and sound isolation and you have to put them in your ears too. Go infection go!
Suno's V4 model has a distinct and consistent kinda rattling laser sound artifact that appears all the time. It's very slightly noticeable on V3.5 but very obvious in 4 so it seems their model is converging on it. I'm curious if that's what's being picked up that got your detection rate to 100%.
Modern pop artists will probably start adding the "shimmer" to their songs to please the stuffed ears of "pop listeners". A bit like what happened with autotune, which is now the mark of a real pro singer instead of being perceived as cheating.
@@mal2ksc I assumed it was a similar thing to how the early GenAI art looked like an acid trip and full of fractals. Like, the weird resonance was the audio equivalent of those artifacts.
I did not know what I signed up for when I started watching you from that noise video I watched months ago, now having watched most of your videos and knowing that you chose to stay with UA-cam even with the offer from companies to what I assume is to make content for those platforms is truly king behaviour.
First time coming across your channel and I am impressed by the quality of your work, research and using tech! Nothing better than mixing programming/tech with music production!
Regarding your AI music detector: You seem to emphasize the difference between music that is created with AI being low res (128kbps) while many human songs are available in lossless formaat. Are all samples in the same quality (128kbps)? Otherwise, what you have essentially created is not an AI music detector but a fairly trivial quality discriminator. I think there's a significant difference between the two.
It’s not looking for the quality of the music. It’s looking for artifacts of compression in higher quality audio. Basically, since the vast majority of streaming services provide compressed audio, the AI models think that most music should sound like it’s been compressed, even when it’s generating audio of higher quality. If you can detect this misplaced or ‘wrong’ compression, you can detect AI generated audio. As a sidenote, you can do the exact same thing with AI generated images, looking for jpeg compression in PNGs, etc.
@@manavali92 I understand - that is exactly my point, that the proclaimed 'AI Music detector' is essentially a quality discriminator. It's useful and I appreciate Benn's efforts here, but we should be honest in acknowledging that the use of compression artefacts to detect pieces of AI-generated music is quite trivial and not a robust method in the long run, given this is not that hard to bypass. That's why I argued he should claim that he's using AI to detect low-quality audio or compression artefacts. The fact that it is an effective discriminator for AI music is nice in practice, but not robust - I know this nuance is much less catchy to the average UA-cam scroller than 'Using AI To Detect AI Music', I get it.
Also, how to spot the difference between an MP3 and uncompressed WAV. It's not a matter of detail or clarity, it's artifacts you need to listen for. The hiss of a cymbal retaining it's metallic sheen and not sounding washy. The echo, room noise, and reverb immediately following percussive sounds and how full it sounds and how big the room the instrument is in sounds. They are very subtle things but in those double blind tests you can get >80% accuracy picking the uncompressed audio if you listen for the artifacts.
I've seen some of your videos before, but it wasn't until I saw your "drop" video with Mr. Bill today that I learned you're also The Flashbulb, and I've been listening to your music for years!
Just the conclusion of your video is a freaking great representation of why I follow Venus and you and will continue to do so... you guys are role models for me !
between your video and GamersNexus's commentary today about recent AI news (eg social media companies standing up their own AI bots), I have to say I have not felt more bleak abt the future of AI. humanity trapped paying for endless gray slop created on ever-thinning margins for - being extremely honest - the sole purpose of preserving the egos of the most incompetent, cowardly, and unimaginative corporate management teams in history.
Benn, please, DO something to organize content creators to fight against UA-cam's anti-fair use bias. Fair Use and Copyright are both entitled to the same legal protections, but every time there's a question about it, UA-cam instantly defers to the copyright claimant instead of defending Fair Use. This has got to be illegal.
It isn't, because fair use is a defense you can use in court against alleged copyright infringement. Nobody can say whether or not something is fair use, except a judge. UA-cam occasionally does side with the creator, but in most cases, it isn't up to them to make that decision.
@@jansenart0 "Fair use is a doctrine in United States law that permits limited use of copyrighted material without having to first acquire permission from the copyright holder. Fair use is one of the limitations to copyright intended to balance the interests of copyright holders with the public interest in the wider distribution and use of creative works by allowing as a defense to copyright infringement claims certain limited uses that might otherwise be considered infringement." A *defense.*
Beatport claim: "In 2022, Beatport paid an average of $0.10808333 per stream, while its service aimed at open-format DJs, Beatsource, averaged $ 0.17773333 per stream - that’s more than 30 times the industry-wide blended average streaming rates across platforms in the United States was $.0053 per stream." Checking my own royalty statements I see payments of about £0.02 per stream from Beatport.
Beatport is a horrible site especially for buying music. They literally have limited the downloads to music you bought unless you're subscribed to their streaming service. They also require you to pay extra if you want to buy the song as lossless like FLAC or WAV which also makes zero sense. In Bandcamp I can literally download the music I bought in any format I want how many times I want with no extra cost.
Hi, Benn! Awesome video, as always: tons of research, tons of insight only someone with your experience could provide and a fun watch. I hope you finally found a greyhound with a big blue rubber deek, and I hope you release another Acidwolf album. ❤
Most everyday consumers of digital entertainment don’t care whether a piece is created by AI or a human. They simply want a song that resonates with them. Whichever song delivers that connection wins. People often forget that digital entertainment was originally meant to provide affordable, accessible content for the masses, and AI-generated, targeted songs are just the next step in that evolution. One song = one listener - this is the future. Meanwhile, another group of consumers seeks a deeper bond with an artist, going as far as paying $3,000 for concert tickets to experience a personal connection of a different kind. Digital entertainment is undergoing a seismic shift - one that many fail to see, distracted by fear, anxiety, and anger toward machine learning and AI. We will have no other option but to accept this reality because this is what the masses want. The best thing talented artists can do is to start educating themselves in machine learning or working with ML engineers on a new set of products that will deliver the connection I mentioned in the beginning on a more profound and much larger scale. You could also start touring and playing in clubs, where audiences will soon crave that genuine, real-world connection.
Except that far fewer people are going to spontaneously discover and listen to that music. And the artist won't be able to make revenue from advertising, since they're a little website probably won't have any advertisers. And it's harder to get people to subscribe through a website than getting them to subscribe to a huge music service.
I’m gonna say this again: libraries of many countries already have sort of streaming services and I think that’s who should be taking up the baton. BTW: my problem with every mp3 versus WAV test I’ve ever seen or read is that they mostly don’t do it with music that I think I’d hear the difference on or indeed that I would want to listen to. I can definitely hear the difference on the live and studio opera CDs I ripped from my own collection about fifteen years ago even though they’re all 320kbps OGG or MP3. Edit: I’m not talking about cymbals and other such high frequency content. I think that’s long ago been taken care of. And I admit that I haven’t tried testing my claims for easily nine years. I’m sure a lot has happened in codec development since then
tbh I only noticed a difference between MP3 and WAV when listening to very sonically crowded music, since that is where the compression hits hardest. Since then I'm an enjoyer of lossless.
I agree. I probably couldn't tell and wouldn't care listening to Michael Jackson, but play me some chamber music, some acoustic jazz, or even the high definition Dark Side of the Moon and I can tell you. I also have pretty high quality audio equipment because I DO notice.
I mostly only listen to metal, and its pretty easy to spot a low bitrate song, the cymbals and hi-hat are a dead giveaways if the bitrate is very low. Just sounds awfully crushed and hissing.
I say that I can tell but even Ben admitted that at 256kbps he lost that ability. I would love to see a video of all types of people from pros to musicians to younger and older to audiophile to average person try to define Flac/Wav vs mp3/opus/aac/oogVorbis. For me, specifically hihats and cymbals tend to lose that shimmery sound. That is normally the giveaway. Could it be that older folks with their hearing slowly getting more and more low passy might have a harder time identifying format? If that is true there would likely be a peak where someone's ear is trained more and they still have not had that decline start. Something like a sports player with the experience and joints all still intact...
It was great to hear on your views on lossless audio formats. Personally, i do notice a difference with formats sometimes...it definitely depends on how the track was mastered. If the track was designed with lossless formats and audiophile grade gear in mind, thats where you hear the difference. Once you get used to a good setup, you do start to notice the drawbacks with mp3 files vs lossless.
You are, without a doubt, one of the most interesting and diverse UA-camrs I know of. You seem highly intelligent, and you're able to break down complex topics and present them in a way that makes them understandable and entertaining. Keep going, please!
The question of which service pays artists the most per stream, is a nonsense distraction. It ignores that the revenue of the artist has more to do with which service will spontaneously increase their listenership, more than the hard number that is paid.
@@braintrust12 Unfortunately for you, the algorithm will prioritize the streaming services' own mass-produced background music. Or maybe they will just add a couple of millions extra listens on their friend's tracks, as Jay-Z did for Tidal.
@@Mnnvint the algorithm cant make people voluntarily listen to ai music. because ai music sucks. if it was good, then sure the algorithm could get people to listen to it. but it sucks so nobody is listening to it
@@braintrust12 Yes, the algorithm can make people listen to ai music, because a lot of people just put a spotify radio on repeat, and ai music can be "good" in the sense that it sounds like the other music it's presented along with if you're not actively listening.
@@Mnnvint i just don't see it. i've never heard an ai song that is good enough to blend in with other professional artists' work. i say "professional" because i've heard stuff from amateur/hobby producers/writers that sounds so bad that I literally think to myself "wow... this sounds like ai made it." in fact, that's the only purpose ai music serves atm: a benchmark by which "terribleness" can be measured.
I kinda want to get into the visual art version of what you're doing as an film animation student. I also have some sound design under my belt and am learning about sound/music: it's a lot of work! People deserved to get paid for their work! Props for all the work you're doing and sharing. :)
I was once in a party with a paid employee of a local bank, which is also a huge IT company branching out into different things, AI included. She worked in their music AI department, she said she was in charge of training data. When I asked her what's the process for getting the music, she said "Torrents, lol". I gave her the coldest shoulder I could. But I doubt there's going to be a comeuppance for this, unfortunately.
Comment FAQ:
Generative AI defense comments:
Arguing that your Suno/Udio music is equal to my music was already done in 2023 when a federal judge decided that AI generated content isn't copyrightable nor qualifies for the same protections as human-made content. If you have a valid argument for this, then go appeal that decision. This is not the place to argue about this. Go read the actual, concise, deciding argument by searching "Case 1:22-cv-01564-BAH".
My "ai detector" model use:
I've already had some calls with data-certifiers/content-ID companies that I had previous association with and it is completely possible to integrate this AI detection into their APIs. What I would much rather see happen is distributors release a blanket policy stating that they will not be paying royalties on content generated on Suno/Udio/etc and directly uploaded to their distribution, and will remove/chargeback royalties earned from that material.
Direct link to case if anyone else was annoyed by UA-cam not letting you copy and paste on mobile: www.copyright.gov/ai/docs/district-court-decision-affirming-refusal-of-registration.pdf
hi, your points on how much the streaming services pay per stream got me interested: what's the pay on bandcamp like?
i'm sure less people buy music there, but a single album purchase must be worth hundreds if not thousands of streams, right?
@@BennJordan "For the foregoing reasons, defendants are correct that the Copyright Office acted properly in denying copyright registration for a work created absent any human involvement."
Yes, for THIS specific case, not for AI music in general involving some human involvement.
You just cherry picked what you want: the reality about copyrighting music made with AI is different from what you think it is.
Again: do you really think Suno terms of use are illegal ? Prove me wrong.
If people enjoy listening to AI and can’t tell the difference between a human artist why shouldn’t they get paid. Someone still has to do the work of generating a “song”.
@@NicolasPerrier Suno is being sued by RIAA because their terms are very much not aligned with copyright law.
As a graphic designer AI art has been absolutely unbearable and really affected how i work, i've been forced to use this chrome extension called 'Pre-AI Search' which literally just filters to searches before 2023 (lol) and its the only way i can actually get non-AI reference designs anymore. Can't even imagine what the internet will be like in a few years.
it especially sucks when it's one of the first few image results...
You can do this on google by putting "before:2023" on the end of a search query. You can try it with the word "skibidi" to see instant proof
If you're on a different browser or don't want an extension, before:(year) works in Google to show images before a certain date, I.e. (your search) before:2022
My graphic design work went from 60% of my work to less than 10% in 2024, the hit is real. Thankfully as a long-time DJ (since the 90s) i have been able to take on more gigs to cover that (don't get me started on the state DJing is in the last few years hahaa).
Welcome to why I hate tech bros
You can detect AI-generated artwork basically the same way. Even when you download AI-generated art as what is supposed to be a lossless PNG file, it still has JPEG compression artifacts in it, because the data it's been trained on is virtually all from compressed JPEG images, so that's what it thinks the images are supposed to look like, and thus the AI model reproduces the compression artifacts as part of its own art.
But hardly anyone ever views a lossless PNG, so the moment it gets converted to a lossy file is it still detectable?
How's that working out for detecting AI writing? These tools are pseudo science bs that only make flawed comparisons.
@@agentbananauk if a file is converted from PNG to JPEG, and then back to PNG, it will have JPEG artifacts. Most JPEG artifacts aren't very noticeable, so it's unlikely to cause problems, but it means you can't just say "this PNG file has JPEG artifacts, it's AI!!!!!", because it could've been converted from JPEG to PNG along the way.
The only way for this advice to work is if you get the file directly from the "artist" and it has those artifacts, or if the "artist" uploaded the image to a site that doesn't convert to JPEG (Twitter/X does convert, so not there)
... and if you just add a jpeg compressed stock image - or parts of it - to your own hi res artwork. How would that be different?
Most of the AI detection tools that try to do anything remotely like this have abyssal accuracy rates, and those success rates are generally higher with older, outdated generative models.
That includes both false positives as well as false negatives. And the moment any editing is done to the image (cropping, resizing, transforms, filters, upscale, etc) their accuracy drops even lower.
My biggest issue with users of GenAI is hypocrisy. I know programmers who will happily use Suno/Udio while simultaneously railing against coding assistants like Cursor which are trained on tons of copyrighted data. I know musicians who happily use Midjourney/Stable Diffusion to generate images while simultaneously cursing Suno/Udio. You don't get to cherrypick which form of GenAI is morally acceptable. Either they *all* are, or none of them are.
Agreed. And all are great tools. The training must become ethical by using dedicated, paid for, high quality data.
Hypocrisy? I thought it was the stealing.
Agreed, 'I am' mediocre in so many things I could not do before. I can scrape a website with python or make some generic image with stable diffusion.
Im not really mad at it. The sad part is how people respond. It's like being mad about some cookie cutter pop artist. It's sad that people love that, not the thing itself.
Udio/suno are neat but I have not heard anything that I would let exist as it does. I mean, it lacks that original feeling.
That said, if your workflow was to generate something there and then chop it up, that's kinda cool.
Then we come to where I actually get excited about this stuff. Stem separation! Freaking magical! The best thing about AI for me. Then, go on HF, and there are like 40000 niche things and Image detection/identification, background removal, then really crazy specific things
You can just use ethically sourced GenAi and the problems go away
To me it depends on the usage.
I don't think it's unmoral for a musician to use genAI to generate a cover image for an album, if the music was recorded by them. The product is the music they made, not the cover image.
If AI generations were respectable you wouldn't see so many people trying to launder them. Everyone knows they have to hide the fact that their so-called "work" is AI.
Once again: there isn't a single legitimate, ethical or benign use for so called "AI" (which isn't "intelligence at all).
I think it is a tool that can be used in combination with human creativity and knowledge, it doesnt have to entirely replace it or be devoid of human soul or effort.
I write all my own lyrics and draw all associated art for my music- album cover, profile pics/ banners etc with no AI assistance.
I use AI mostly to generate vocals which i can control the pronunciation and flow of.
All I want is some appreciation for my lyrics and AI allows me to present them as a fully finished song which I have a ton of control over.
its the best medium I currently have access to be able to share my music right now, so I am grateful for that, and i am proud of what I made.
I cannot perform vocals where I currently live due to shitty circumstances and environmental limitations.
I use AI and try to retain as much authenticity as i can to share my work, and i think i accomplish that skillfully, better than any other AI music ive heard thusfar.
@@ITZTLI-music Dude, then make instrumental music. Like I did when I couldn't sing. If the issue is only technical and not that you actually cannot sing there are many many ways to solve with without resorting to so called AI
@@ITZTLI-music me using Microsoft SAM
@@ITZTLI-music so you're supposed to find someone to do the vocals, not find an app that is using someone else's vocals without their knowledge or compensation
This stuff makes me super sad. All of a sudden on Spotify my related artists went from a bunch of artists in my general space to being all AI artists and it just makes me feel like shit. Discovery was already horrible on Spotify and now it's just impossible.
DONT lie Neon Lights be off da Chaaaaaaaaiiiinnnn dj Shooppuiiiuhhgggfdx wooo
For me, it's just been stuffing pop music into my fov and playlists. I am almost exclusively a metal head with obscure tastes. Incredibly frustrating.
They changed the fans also liked algo it’s now garbage . And puts u with irrelevant artists
I've always used UA-cam for music because my taste is someone niche, but even it is getting overrun with AI music. Trying to find a playlist of genuine, non-AI classical music is way more difficult than it has any right to be. And "classical" AI pisses me off even more than others, just like the ones that rip off renaissance paintings and try to pass them off as genuine. As if grammar and spelling aren't already at an all-time low with the internet; our knowledge of art & history is damn near done for if this shit isn't fixed.
@@courtykatclassical ai brb
6:58 "There's nothing fair about stealing an artist's life's work", says the RECORD INDUSTRY, without a hint of irony
To your statement, the record company execs would reply, "Hey, we didn't steal anything! We swindled, it, fair and square!"
Then don't sign with them. Market it yourself.
@@TheNashEquil lol
Then don't take their money
lmao who invited the record company simps?
The low pay per stream combined with the growing low quality content has moved me towards alternatives. I don't stream anymore and instead support my favorite artists on Patreon and I buy music from the artist (often on medium like disk or tape) on Bandcamp. I feel better knowing the artist at least earns something and I much more enjoy the music I listen to. Remember that time you went to the music store to buy a CD for that one song and you kinda start to appreciate the rest of the album too? Well: that feeling. Please keep making these videos, they are educational and a joy to watch. Thanks.
I always buy FLAC (if they have it) or 320KBPS MP3s of what I like, and I download them right away and put them on a thumb drive. I don't feel like I own what I bought until I can listen to it without a website giving permission.
@@FourWaveforms"I always buy FLAC (if they have it) or 320KBPS MP3s of what I like, and I download them right away and put them on a thumb drive" - For anyone who doesn't follow this advice, definitely start doing so, especially on Bandcamp. Do NOT assume that because Bandcamp has an app with which you can listen to your library, that it will stay there forever. Artists can, and do, remove their music from Bandcamp (or remove individual albums, or individual tracks from albums), and when they do, it's removed from your library. If you didn't download it when you had the chance, then it's gone forever.
That certainly a good alternative. However, I think artists earning money from art will be going away. Being able to scale earning from your art quite modern, and I honestly think that making art for art's sake (instead of for likes or money), is the core of art, and I don't think art it will go away even if nobody listens or pays. Bach or Kafka would still make their works.
@@TwilightThrone And nobody will ever listen or find that art. Playing to an audience of tens blows.
@TwilightThrone but Bach was a church musician who made money from performing. I dont think if he was preoccupied with other work that he could have been as productive as he was. The way I know that is that almost none of the most famous composers in the history of the western musical canon worked with something other than music as thier career. It will be a massive loss when no person can survive on thier art alone.
Using the concept of missing song data from file compression to catch AI is FRIGGIN GENIUS !
You're officially the "John Connor" in this geeked out robot battle. 😊
I'd love to have a browser plugin or something to get to use that AI for myself. I've seen SO MANY AI songs that aren't caught in the playlist linked below the video (because there are probably hundreds uploaded each day?).
Also, I didn't understand how the "fingerprint" was created to find AI music with 100% certainty; even though I somewhat know how LLMs and other types of AI are created.
It 100% is pure distilled genius.
i saw a similar thing of catching ai generated images. ais incorporate so many compressed images into their training data, they mimic a lot of compression artifacts in their outputs that arent natural or dont make sense when they mash together different compressed images in the training set. because these compressed files are so baked into the training set, it doesn't seem like it is possible to remove these artifacts from ai models unless they were to retrain their models on only uncompressed files.
I don't get how it doesn't just catch all mp3s?
Longtime Qobuz user here. Glad to hear they pay a little better (though as you say they don't have the volume as they're a niche player). But the other main reason I like them, aside from the excellent sound quality, is that you can purchase the music, either tracks or whole albums, and get digital downloads (FLAC, mp3, etc.), DRM-free, to keep. And I imagine those pay artists better than the streaming. So I regularly buy music on there that I'm finding myself listening to so I know it won't disappear, and hopefully compensating the artist.
Yeah also a big qobuz fan here, I've found some music on their site that I just couldn't find anywhere else and bought it, paying the artists big bux and getting some high quality shit. I fw it🔥
What helped me choose Qobuz briefly, was that Qobuz pay per view being the highest. Tidal lets you give more money if you’re on their highest plan to your most watched artists.
Is this an actually ethical streaming service? Never thought I'd see the day.
not to mention that, unlike Tidal, they have FLAC for most of their catalogue. (Tidal uses MQA for much of its catalogue, has been called out on this, but never did anything about it, making them commit false advertizing)
YT is already loaded with low effort AI slop by people who are just click farming. The LoFi genre especially is plagued with it.
Yeah, I don't think they actually pay for streams at all, do they?
@NicStage, Ad revenue just like any other YT video. You can fit a good number of ads in a 55 minute “mix” of AI music.
BS.. Neon lights neon remix v 2.2.21212111212 Is Lit. Neon
@@NicStage ai music
i joined an ai music group to spy on people and yeah, these guys are posting their music on spotify and trying to claim themselves as artists and music producers. some even sell their ai songs as legitimate beats to real people. some have gotten in trouble and their ai songs have been taken off spotify, but they legitimately see themselves as artists. ive tried to listen to a few but they all sound like trash. left the group because i couldnt stand the entitlement of these people.
This video is going to have a whole ton of people terrified of being caught out
Expose them all.
Caught with what? loops have been a thing for ages now, its in the video.
Spotify is actively promoting AI generated music because they can create their own content and not pay musicians.
@@KwissBeats Did you watch the video? AI created music is not copyrightable in the USA. People are taking unaltered AI generated music and committing fraud by uploading it as their own creative work to establish a copyright claim and receive revenue. Artists that do this are not "artists" and are lying to the public. Separately, calling music generated with loops you own or possess a license to is not fraud, is copyrightable (for your work but you may not exclusively claim copyright to the unaltered loop) and Benn points that out as the "Splice Problem" Venus Theory originally named. Benn clearly says the problem with loops is the content ID system with media platforms attaches copyright exclusive ownership of the loop (which is not correct) to the first user on their platform with the registration and neither the loop publishers or media platforms can be bothered to implement the simple, but not free solutions he illustrates.
Gooood.
As a Data Scientist I love 100% accuracy
Same. Like do people not have the time or talent these days to learn an instrument? 😂
🤖... I like human fluidity.
As a random person on the internet; I approve this message
😊
As someone making ai for a living the past 6 years, 100% accuracy is at best an indicator of something gone way wrong in methodology. I like your content Benn but be sure you are not advertising something over confidently! I know I have in the past. Data leakages and not using good test/val datasets happen to all of us
That being said, if there is actually a signature in ai audio that is 100% distinct from the rest, then I really admire that you found it! Maybe cnn are overpowered for the task too :)
1:00 wait until you find out that spotify does this themselves then jams it into playlists where people arent going to be paying attention
Turn off Smart Shuffle and play only music you added to a playlist by yourself.
@@mrolowu1 so that's what it is about... I tried it and it just started adding random junk I would never listen to. I dont like lyrics - it gives me lyrics. Why?
Not true, they use real artists and work with labels to "jam" the playlists. It's for producers who just want to make music and earn a living.
@@koppy82 ....have you not seen the endless amount of videos on SPOTIFY'S AI playlists? These are spotify curated playlists.
@@koppy82 it's literally already been proven, I'm not saying what you said isn't happening but they absolutely are publishing ai music on their platform themselves
I appreciate all the people on a music channel straight up coming out and stating they have absolutely no love for music, art as a whole, or the artists who create it.
Nobody said that. You just decided people said that
Elaborate or else
So stoked that a bit of my music was used in this video at 4:40 thanks Benn you rock! ❤ Appreciate your content very much 🙏
You're doing some of the best journalism online, Benn. And you're doing some of the most effective science communication online, as well.
I haven't finished the video so I apologize if this is mentioned, but one thing the GenAI boom has also facilitated is the age-old tradition of buying listens/likes/views/etc. A while back a friend - who is in a fairly successful band - and I were talking about another band from our scene that got signed to a major label before ever playing their first show. Now, obviously they had connections, but then the label has to make it look legit. My friend was like "oh yeah, here's what they do," then pulled out his phone and pulled up the band's Spotify page - all of their suggested "similar artists" that fans "also listen to" were AI-generated "bands" from Eastern Europe that only had one song each, each with tens of millions of "streams." He explained that you can just rig the whole thing with AI, buying millions of streams without anyone ever having to actually listen to anything, but there will always be signs when an artist does that.
how is buying listens “ai?” lol people were buying listens 20 years ago on myspace
Capitalists just filling the world with garbage to make a quick buck. And odds are that band will just be chewed up and spit out and it’s on to the next one. Fucking vampires. I can’t see being the artist to agree to that deal, but I really can’t imagine being the guy at the record label in charge of creating fake credibility for the new act that just got signed. You can’t take enough showers to wash that off.
Your content is a (good) outlier in the world of synth/music UA-camrs. Even if you think some of your material is too nerdy, complex, or weird for standard audiences, I encourage you to follow what you love.
Thankful that my taste in music is so pretentious that AI hasn’t really broken into the genres I listen to… yet…
It is a little misleading to say that compression throws out, say, 80% of the INFORMATION. It throws out 80% of the DATA. In theory you can compress a one-hour recording of a sine wave using two numbers, the frequency and the duration: e.g. sin@440/3600. That is WAY less information than a 128kbps hour long recording, but the reproduction would be BETTER than a transmitted recording. All sound can be reproduced as sine wave combinations. So, while clearly music requires more than one formula, compression does not PER SE mean information loss, just data reduction. Further, sampling error forgiveness is not JUST a psychological effect. For example, so long as the sampling rate is twice the highest frequency, Nyquist and Shannon showed that there is a mathematical stability to a sampled result--a result that can lead to compression mechanisms that do lose information, but not at the same rate as they lose data, for example by first filtering out barely audible high frequency content.
Great clarification about data vs. information. To say a bit more about the Nyquist-Shannon sampling theorem: while it guarantees that frequencies below half the sampling rate can be faithfully reconstructed *in theory* -- the practical quality of frequency reproduction depends on factors like bit depth, filter design, and other implementation details.
But that's not how Mp3 or any similar lossy compression system works. It absolutely throws out information. Sure, it isn't necessarily information you would have heard so you could say, in a sense, that is was redundant, but that doesn't mean it wasn't information. If it was a graph of share price, for example, rather than music, then you wouldn't say that throwing away 80% of the data did not lose a large amount of information. If the information was still there, you wouldn't be able to do a delta between lossless and lossy compression.
Lossy compression clearly throws out information. Just encode anything on the higher frequencies of the audible range, compress, decompress, decode back to regular frequencies, and you'll see how much worse it is.
Storage is so cheap, lossy compression is ridiculous.
Great explanation. Another way to think about this is JPG vs SVG. A rasterized version of a vector image might be 5MB, but it's vector original only 50KB. The vector image represents the same contents in different way, leading to similar, or even better reproduction at higher zoom levels.
Apparently jpeg artifacts are a problem with image generation, too.
My guess for most profitable platform for artists was Bandcamp, but that's a fundamentally different model.
and people will cry and say "we CAN'T just use bandcamp!! we just CAN'T!! It's too hard, spotify's easy, waaahhhh!!"
I mean... You can stream from it x)
But only the albums you bought to begin with.
Similarly true for dizzylab, which is a Chinese "replacement" for Bandcamp since BC got banned a couple years ago.
Probably no, unfortunately. It's technically one of the worst as you can listen to an artist for hours without the artist earning any money from it.
I personally don't use any paying steaming service and listen to my music primarily on Bandcamp and on UA-cam. I've bought some music on Bandcamp, but just a tiny fraction of what I've listened to more than once because 5-7€+ is pretty engaging.
It would be interesting to know, for The Flashbulb for instance, the approximate pay per stream if we compare the number of listens of 2024 to the total of purchases over the year. I have really no idea how many people on Bandcamp actually buy their music and if that model is working well for the artists.
As you say, it's a very different model, Bandcamp being primarily a retail store which you can use as a streaming service too. But perhaps it shouldn't stay that way.
I would personally gladly pay 10€ a month to Bandcamp to have say 9€ being distributed to all artists I've listened to during that month. They are perhaps afraid of going in that direction, idk.
I was gonna guess Deezer
i like bandcamp....
hey Benn going off the happy note of the end of the video, I did want to say, I have recently canceled my Spotify subscription and have no intention of going back. And your videos on Spotify and how they treat their artists were the main reason I did so. In its absence I've been buying most of my music off iTunes which might not be the best as a large portion if not all of the money may just be going to the labels, I am still happy to have made the switch off Spotify and will keep trying to find ways to more directly support the artists I love. thanks !
Disappointed in many of the comments on this video. Where is the beauty in buying creativity from a company of thousands of engineers rather than cultivating it on our own? No one can convince me that I'd be better off with this garbage product than with my own hands. I don't care if I don't produce anything worthwhile, at least it's me producing it and not some amalgamation of stolen material.
Think of it this way. If you, like me, have shared a talent (musical or not) with another person, you might have received a comment like, "That's so cool! I wish I could do that/I always sucked at that/I'm not smart enough to do that/etc."
These people usually genuinely believe these limitations are something we were able to surpass via a stroke of fate or a blessing of some sort, or that these limitations are simply given only to some people and are not surmountable. Alternatively they're self aware enough to know that with practice they could probably get to our skill level, too.
And the biggest thing to remember is that that desire is legitimate and often even stronger than you'd first think. But many obstacles stand in the way and they're many of the same ones we all start with when learning a new skill: fear, lack of energy, lack of willpower/ability to care enough, low self-confidence, sacrificing time, spending money, and others for certain.
But what if you could hack off a lot of those obstacles? Well, what if the odds of winning the lottery were suddenly raised by 10,000%? A lot more people would win the jackpot.
From there, it becomes a matter of values and ethics. Artists value artists. Not everybody, though, values artists, or at least values them enough to have those strong ethics that give the person the desire or willpower to refrain from using things like generative AI, pirating independent artists' music and small artists' art/films/writing/games/etc.
These tools are truly appealing to many people. So what gives a person reason to choose NOT to use them, let alone advocate against them? As a society we don't generally respect artists the way we should. This is evident enough in the art/media industries/economies.
@@saltiestsiren💯💯💯💯💯
@@saltiestsirenthey will miss us when they live in pods looking at the same animation to the same sound cue for the 100,000th time
Lossless and Compressed are not mutually exclusive. It would be more correct to use the terms lossy and lossless to differentiate. FLAC (lossless compressed audio codec) is identical to WAV PCM (lossless uncompressed audio format).
was looking for this comment, cheers
And almost all people wouldn't be able to tell the difference between that and it run through a well built compression algorithm. That's the point of "Lossless"
@@brianbauer3148 the only way I can tell an MP3s isn't compressed well is if the cymbals/hi-hat sound wrong. But I have some 192kbps MP3s that sound just fine
@@brianbauer3148 why is lossless in quotation marks, it's literally lossless, it's lossy compression that's built on the idea that u can't tell the difference anyway
there is also the possibility that the experiment they mentioned had flaws. we dont know if the equipment and song choice and people had any affect on the sounds. plus the song itself could be lossless but still have lossy samples . and some people like myself whom are sensitive to sounds can really tell when audio sounds off. i for one cant stand lossy audio. at least when its compressed past a certain point. the diffrence is there but i think its not "most people dont notice" likely its "most people havent experienced the difference and are used to compressed audio as the norm" i mean most of the audio consumed nowadays is trough compressed social media sources
The Splice problem is not just a sample site problem, it can be a synth sound patch problem as well. A friend made a track using a preset on Reaktor. When I heard it , I asked him why did he decide to sample a certain artist. He had no idea what I was talking about. He then showed me the preset he used. I then played the track (a pretty popular track) and we both were floored. UA-cam would’ve definitely flagged him even though it’s a preset sound and not a sample.
Starsky Carr has several videos addressing this where he attempts to demonstrate filter sweeps on analog synths and they get copyright struck by someone who has taken Moogs, Oberheims, Prophets, etc. and just swept the filter slowly. That kind of BS can definitely threaten UA-cam as a platform for gear reviews. UA-cam gear reviews have plenty of problems (lack of transparency of financial support to reviewers for one) but taking a raw saw wave and sweeping a low pass filter ought to be an easy "non-copyrightable" sound. Sweeping a filter over a raw wave doesn't meet anyone's definition (except UA-cam) of a "transformation."
@@scott21113i mean i guess? but more importantly, youtube isn’t “threatened” by anything. they have a 100% monopoly lol
@@scott21113 META is adding ai and bots to their platform. AI has been slowly increasing in frequency in my feed every couple of months across ALL platforms. If the system doesn't collapse under itself or the AI becomes self aware and destroys their own makers a la Skynet, I don't foresee anything happening. The big record labels don't have the balls to take on Google, Meta, and ETC who are accelerating the rot. First Mistress Orange Trump assumes office soon with Lord Elon wagging his money collar AROUND THE WORLD putting his short nubby fingers in every sovereign nations "honey pot."
Have you noticed so many platforms have began sucking off these psychopaths the last couple of years trying to get in the good graces of the hopeful plants they paid for on the campaign trail?
AI has and will continue to be a massive mistake, but the pandoras box is open. The west will fall before it regulates itself to ensure it doesn't self immolate, its practically THE American pastime of voting against our own interests. I would be shocked if the shareholder suits banded together to right the ship they've set to sink slowly over the last decade.
It should be noted that streaming revenue differs vastly by country and subscription status. Spotify free from Argentina pays 30x less than a paying subscriber from Switzerland. Also, UA-cam Red from Singapore pays very well, about the same amount as Amazon Music in USA. My dataset is pretty small, but overall it checks out with what Benn said. Spotify genereally pays about 0.0018 from countries other than the 10 richest western countries.
I failed to install any Chinese streaming apps, but some of the music that got submitted to them just disappeared. (Again: I'm a no name brand musician that made a few ambient songs so probably just got flagged as spammy meditation music and blocked by their automated process, and not because of some evil scheme to steal my half a cent per year.)
Funnily enough, the sole exception is russian VK. They pay pennies to their own content creators, music, video - all the same. Just a pointless platform
it also differs by subscriber type. pelaton users only listen to their streaming platform while on their bikes. this severely limits how many total streams users have. divide the already low streams by the high monthly cost and you get a very expected significantly higher than normal pay per stream. the metric is just dumb. it’s used by people who want clicks on their video/article/whatever
Your videos are extremely relaxing and thats probably my favorite thing about them. I think its because you passionately speak about what youre talking about, but also that you keep it entertaining and smart enough to invoke interest and learn from. Its like an ASMR video that helps me learn about music production
Such an elegant solution for identifying if something was created artificially.
Just getting into your videos. Not a producer, but programmer. This scratched my brain in a new, awesome way
Copyright and patents need to be completely reformed.
Noo just Add More neo lights
100%
look up their history and realize that the only reason they exist is to shield a few from competition.. they are constructs to to allow a few to maximize profits without the market being able to correct this, as competition is prevented from doing the same thing. That is the definition of a monopol.
@@joansparky4439 Patents were designed to promote R&D, so others wouldn't copy your 5 years of research the day you make them public through a product. Similar for copyright, you don't wanna write your masterpiece and have hundreds playing it without you seeing a dime.
That said, these reasonable ideas got lobbied into a monster that keeps corporations profiting off dead men for decades after they're gone, while holding culture and technology back from running with these new ideas for entire generations.
Nah using AI is pathetically lazy and you aren't an "artist" using it
19:16 wait, your what?
His peaness, the characteristic of literally being made of peas, which is at the lowest state possible due to him being a real human being.
I was thinking of giving this to a class, and that'll be a good test of whether any of the students watch till the end.
lolol
You heard him!
That really threw me. Wtf dude!
This video is what’s up! Another thoughtful and relevant piece. You’ve convinced me to both sign up for your Patreon and, push to to inspiration from your work to make more meaningful content on my own channel. Long time fan Ben, keep it up.
Just discovered your UA-cam channel today, this is my second video and I realized oh shit, you’re flashbulb! Been listening to your work for a while now
Love it that Benn and Venus Theory had a talk one day and they both decided to start giving zero fucks and being way more honest and freewheeling in their videos. Love it.
Benn, what's your plan with your AI music detector ?
Will it become a public tool in the future so people can identify who's using AI ?
I've shared it with some colleagues who are ML engineers. Ideally it would be used by services like BMAT.
@@BennJordan Are you willing to share any details about model/method, training dataset size and number of epochs, validation set size, etc? A 100% accuracy result on a classification model is generally a very bad sign, and what I could find from the academic community (mostly audio deepfake detection, not necessarily music) was in the 90-95% range.
@@spacepxl Yeah, for classification models 90-95% is good, 96-98% would be brilliant, 99-100% is at least concerning. I agree with you: it would be nice if they share with us the pipeline in the future to replicate the model using our own training data
@spacepxl honestly, I thought even putting that little fingerprint image on the screen was too revealing. I hope he keeps it secret for as long as possible lol
@@BennJordan I think you have to keep in mind that my father's generation was also raised on music and were very involved with Hi-Fi systems. My father to this day isntantly recognizes Mono sound even though I think I have the better ears. But they were raised on these media similar to newer generations being able to tell AI voices apart from real ones much better it's a matter of how early you got into contact with it and how you trained your ears. Unfortunately many people do not consciously listen to music anymore. It has become just another commodity. But there is one huge difference with compressed files and lossless. Moving pants in front of the subwoofer a DJ told me. He always can tell if it's compressed when the air isn't really moving due to the lack of infrasound. And I would bet that if you showed younger people comparisons they would get better over time. That is not to say that they would be able to tell the difference with the higher bit rate. But they surely would be a whole lot better. I surely did.
Interesting about compressed vs lossless audio. I general I have noticed a difference between FLACs from CD audio vs mp3s. Usually the cymbals get distorted first and you start to hear a watery noise as the quality drops, as if you have water stuck in your ear or you're listening from underwater. I've never done a blind study though but I'd be interested, but the difference is there to hear. Unless you've spent time comparing A and B however, you may lack experience to make that judgment. I also suspect there are times where the right bitrate of an mp3 and the right frequencies used make it less susceptible to being analyze in this fashion by a person, I suspect.
It’s like wine tasting, just because most people drink wine and think it all just tastes like grapes doesn’t mean there aren’t flavor profiles one can easily distinguish if they know what to pay attention to. Similarly music listening is a passive activity so most people listening to lossy audio streams just hear “the song” they’ve always heard and liked… it doesn’t mean there aren’t musical layers in the signal being thrown out that a discerning ear can listen for a hear. I can always hear a difference between lossy UA-cam/spotify music and my Qobuz stream and flac files. To be fair, cheap Bluetooth earbuds that a lot of people use are poor equipment to detect these differences. But us music connoisseurs who have resolving enough gear can always hear it. And the kicker is the higher bitrate streams/flac files always sound more emotional/musically engaging to me. I feel closer to the music
I did a non-scientific non-blind test on myself compressing down a ripped cd track (can't remember the genre but it was a good mix and production, before the loudness wars) at 320, 256, 192, 128, and 96 CBR stereo mp3s listening through low-mid range headphones. I recall starting to hear differences between 192 and 128kbps (edited) as the stereo image seemed to narrow. As you say, cymbals seem very susceptable to distortion in lower quality mp3s, and I hate that sound
Regarding the last statement.
I left spotify in 2020 and within the year had bought most of my favorite albums I found on it via Bandcamp Fridays.
To this day I still keep an eye out for it to come around and happily pick up an album or two for the money saved from not paying for a streaming service.
I couldn't recommend it more to anyone else.
And it's mine.
It's there when I want.
I get to enjoy it in all the FLAC glory that my ears are too stupid to hear.
I still buy physical albums in the form of CDs when I have the option to, I rarely actually listen to the CDs but I rip all of them using the FLAC codec. All my bought digital music comes straight from Bandcamp. I just like the idea of owning my music, and giving the most amount of money that I can to the artist. All the physical CDs I have go back years so they also hold a big sentimental value for me.
I could not agree more with both of you.
Bandcamp is the best. You can choose from a variety of file formats. Your music is always there, if you need to re-download it or choose a different file type. Once you buy it, it is yours. No DRM. The meta data is usually flawless, so it shows up correctly on music players, and like you said, it feels great to buy music on Bandcamp Fridays. I can't say enough good things about them.
With that said, it's so frustrating when artists don't put their music on bandcamp, because then I'm forced to go scouring the internet for how to buy a digital album from somewhere other than Apple and Amazon, which is bafflingly hard. I will only buy music from amazon as an absolute last resort. This led to my discovery of 7digital (kind of terrible website and download process) and highresaudio, both of which have limited music libraries. I have had luck finding albums on Qobuz that weren't on bandcamp, as well, which I feel better about seeing that they are the top-paying streaming music site.
I might go the route of buying cds again, but PCs don't even have cd players/burners anymore, which means that I have to buy something external. Streaming has made buying whole digital albums so convoluted. Don't even get me started on digital music players and all the annoying UI flaws that they all have. When streaming became a thing, it's like everyone just collectively gave up on continuing development of the hard drive digital music experience.
The higher quality your speakers are, the more likely you are to notice.
@@pronstorestiffi Same same same.
If not raiding the localish Record shop (Wuxtry) a couple times a year I try to order directly from artists on BC.
The nice thing about buying local is that most of the time you get an included BC code card as well these days which is really nice.
And regarding the "joke" about lossless and compressed, I'm aware and can tell the difference between MP3 compressed and WAV/what have you.. Hell, it drives me insane when making music how I can hear the compression happen from render lol.
It's just that at a certain point it gets silly to me personally as I can't tell the difference. Even after trying some *very* expensive sets for funnies in testing.
My dailies aren't the best being a set of DT990 Pros driven at 250ohm properly. But they're enough to make it obvious I'm listening to UA-cam VS local 99% of the time.
Please create a music streaming platform that is fair and just and I feel that you are the man for this! I would help with whatever I can (UI/UX design, graphic design) to make this happen and I mean it. I think that others would chip in too, it could be a community project for the community. Thanks for all your great work!
Most (or just about all) of us complain about AI in the artist realm from the comfort of our lounge couches. And then there are genius heroes like Benn and Cameron who does the work. I wish I had half of your brain and all your guts Benn. Your videos are SERIOUS game changers - keep the light burning.
19:17 throwing an expletive to check if we listen till the end…
>thinking "penis" is an expletive
That's not an expletive, that's a body part unmentionable in polite Victorian society!
Nice WTF moment, makes you wonder exactly what goes on in Benn's. Patreon content...
@@skierpagenah just community content
You know, . . . I just grab a musical instrument, any instrument, and just start playing. The business of business and the highly-processed product consumption . . . just makes me ill. Hang out with friends, man. Tune in to the stories and rhythms that are happening with each other right now. It's so much richer to be alive than yanked along by faraway sound-hook factories.
You mean anyone can just sing or play music whenever they feel like it? Is that allowed? 😂😂
I really like this. But also some people fell in love with the recording arts. Hi Fi gear. Preamps. Tweaking knobs. Finding mic positions.
Lost in time like tears in the rain...
Not all of us can be an extroverted “people person” who wants to collaborate with other humans. I play real guitars and I’ve played in bands. I currently spend most of my free time designing and building my own custom instruments out of wood and metal. Mostly with my hands. I’ll get something cut by laser or CNC if it makes sense and I’ll use samples generated by AI if it makes sense. I get a lot of joy from the things I create on my own, on my computer while not having to depend on other people or compromise my art by letting others determine what it should do or whether it’s good enough. Nobody gets to tell me that my music is any less valid than the next guy’s. If you don’t like it, don’t listen to it.
Oof. Hard disagree. The amount of enjoyment i get from producing art depends heavily on whether or not others are enjoying it. If im not inspiring somebody or bringing somebody happiness with my art, then im failing as an artist. Mind you, there’s nothing wrong with your motive for art either! Im just so tired of listening to this stale ArT mAn take that insists art has to be some kind of masturbatory practice in order for it to be authentic or real.
Thanks dad that’s super helpful
In a world where “ that was inspiring” is something I think less than I could have ever possibly imagined, I’ve just typed it in a UA-cam comment on a video that inspired me and meant it. Well done. Well done, indeed.
Hey man love your content. I think you should take all of your videos from the past few years and boil them down into one large video explaining all of the information involved with the music streaming scam industry. I dunno. Anyway keep it up!
Benn your channel is so awesome. I am not a "music enthusiast" nor a musician, but I really enjoy the science behind your videos. You do your research and bring a genuine, honest air to whatever subject you're approaching.
Having used Suno for a fair amount of songs, about 500~ or so, I've noticed that in almost every creation there is a weird residual sound that appears in its creations. It sounds like a bleeping synth and it usually occurs in parts where the song is becoming a bit more calm, so right after a chorus or before a buildup. If you hear that sound, you can immediately know that it's AI generated.
instagram already tried integrating an AI detection system and it caused a bunch of people to think artists they loved were using AI when they weren't because AI gets lots of stuff wrong constantly which people will know from how often Google's AI confidently gives a wrong answer
Using AI to find other AI... We live in an anime now.
The "detection system" was literally just reading meta tags from edited files and those tags are only placed there if you use a generative AI feature before saving the file. In literally every case I've seen it's been admitted to eventually, usually with the user saying they "didn't know" that one feature counted as AI. There was no detection algorithm, literally just reading the metadata to see if the file had been tagged. Photoshop only tagged files that were in-fact modified using Gen-AI features. So no, nobody was wrongly accused via that system and anybody who says they were is about as believable as your average video-game cheat who says the same thing. "It wasn't me I swear! The system messed up!" Sure thing.
Found your channel and bought the zoia because of you! Your interests and experiments etc are very inspiring thanks Benn... you're adding massive value out here
I think the work you are doing exposing serious problems in the music copyright system is very important. Hopefully, this will be dealt with fairly, eventually. Content ID has been abused already by bad actors that steal monetization with no downside risk to themselves. Now, with AI music production in mass, it is spinning into a fraud race to the bottom that threatens all creativity.
How has youtube never recommended me this channel before. I already watch stuff like this but its just more and honestly a higher quality than most channel.
okay, ill join your patronite, my first patronite ever
keep up with the good work, i can see that its genuiently brings you joy, props for being independent!!
This is, quite simply, the BEST video on the topic of.... wait, what was the topic again?
detecting AI music
not that I care, I make it
@through-faith-alone "Not that I care, AI make it." There I fixed it for you.
@@Simon_Denmark thanks!
@through-faith-alone No problem, anytime.
I couldn’t tell the difference between compressed and lossless on individual tracks but over time my ears tire from compressed audio. When MP3s came out I loved them until I started to lose intrerest in music. Vinyl and lossless brought my interest back and I can listen all day at home now. In the car on mp3 im good for an hour then it’s podcasts.
Even lossless music in the car is tiring tho. The inside of a car is a terrible acoustic space.
@@sinane.y Care to elaborate? Is it too dead of a space? I can assume speaker placement has to do with it as well, as most are in the door down at foot height and some tweeters far in front, close at the windshield.
@@dustinjaros A lot of it is just having to play the music loud enough to never notice the noise floor of being in a car.
Exactly right, the compressed garbage causes fatigue. Single track a/b tests do not account for this.
I only discovered this channel a month ago, so I appreciate I'm massively late to the party but I don't think its hyperbolic to say I think its some of the best content on this platform for my personal interests. Thoughtful, considered and articulate content from someone who seems like a really top bloke. I'm bingeing the back catalog like a boxset rn.
I decided to buy all the music I actively listen to a while ago. It feels better to know that I’m supporting the artists I enjoy, and to own the music I enjoy. Absolutely loved this video, thanks for doing the research!
I don't have a single music streaming service. I tried free Spotify for a week and went back to my MP3s. Thank god for being such a fan of 90-00s Warp Records and 80s heavy metal. It all fits in my Sony player which has a 3.5mm headphone jack for my 12 dollar headphones.
I used a combination of mp3, bandcamp + pirated Spotify. Not a single penny to those greedy bastards.
Legit👊🏻
I love me an mp3. Even if its crunchy, or glitches from years of data loss being moved around from ipod, to sd card, to USB stick! Its like that "dirtyness" of an old mp3 tells you a history of it. I think high quality wavs and flacs are great for certain songs, too. but i'm more of just a fanatic music hoarder, instead of a full blown audiophile, and those lossless files tend to be huge. i know i sound like a quanity over quality person, but thats just how i love having my music!!! Love the video man, puts an interesting perspective on streaming services as an artist! Great work!
Same! Some tracks can sound even better when ripped in mp3 with a low bitrate.
The other problem with streaming is that everything is focused on single tracks. This means that the depth of an artist does not have a chance to emerge. I still listen to my old vinyls and CDs with pleasure.
What continues to amaze me is how few people realize copyright as a system is fundamentally broken.
This is extremely useful work with lots of applications for a pretty niche field of hobbyists and artists compared to the numbers and views you could pull from chasing a more viral topic with your interests. Thank you for shining a light on how AI has been shaping the monetization of music for artists and people who actually care about the work that goes into music. Even though I only produce music as a hobby its so annoying how people who have a true passion for creating new music can have their work undone by some kid with a keyboard and a subscription fee, built off the backs of thousands of years of artists with visions for their art rather than a chat input.
I have fallen in love with your channel ever since that video about Twitter Bots, lovely
I would rather get that info from someone who actually makes bots, not some random video maker who has never made a bot once in their lives and likely has no idea what TCP even is.
The ability to detect MP3 depends on the high frequency content of the track. MP3 (and others) break down when there is a lot of high frequency detail, since I listen to a lot of live real drum music, if the drummer rides the open hihat or the ride cymbal it creates a flanger or phaser effect in the MP3. Once heard, can't unheard it.
A track with 808 drums can be indistinguishable as MP3 though.
I tend to hear it in cymbals first, too. And reverb tails if it's a sparse enough arrangement.
The best way to support musicians is to buy their music on bandcamp or mirlo and then sync it to your music player like it's 2004, or run Plex on your computer and plexamp on your phone. While that doesn't pay per stream, the 80 cents or so that you pay the musician up front goes a LOT further.
I use Jellyfin (lossless streaming my entire archive) and Syncthing (offline listening with compression, think of it like a podcatcher)*.
I tried to explain to my quite tech savvy friend just a few days ago that my music archive of mostly *lossless* music is expected to not grow beyond 2TB. I'm 40 now, a FLAC album is about 400 MB, if I buy a new album per week (which I don't), 50 in a year for simplicity, until I'm 60, well that's ridiculously little data. Estimates beyond 20 years don't seem to make much sense. But it's safe to say that storing audio is a solved problem today and storage density will still grow.
The compressed version of my archive fits on a finger nail (an inexpensive SD card, the entire library can be compressed from FLAC to Opus in about 70 minutes btw ) today. So I can have everything I *like* always with me. Among that some artists and albums that are not on streaming services for various reasons. I tried to convince him with statistics and data. He didn't get it said "there are 50 million songs and albums (?) on my service". Well I know about GenAI for music, he does not, and I have seen on vacation, when we traveled abroad, that his service contains uploads from other users. Lossy and piracy, and bad tagging if you so will. He does not know what his service has to offer. The amount of inconsistent (user provided) garbage and generated garbage is growing. However telling someone that the service he pays for feeds him mostly garbage while the actual artists benefit the least from a monthly subscription, well that's not winning an argument.
Look, I was a pirate when I was young. I knew that what I did was not legal. I developed a sense for quality and I can now pay for things. Meanwhile it looks like businesses are pretty comfortable with piracy as long as it benefits them and they have their PR mouthpieces touting that it's not piracy and a fair opportunity for everyone.
*I can add an entry to a text file to "publish" something and the album will be downloaded for me (while still having the option to stream with Jellyfin). Yes this even works from my phone. I'm sure that with today's possibilities of coding something the process could be streamlined a lot more. Plex and Jellyfin don't seem to be much more than a web UI for FFmpeg. It's just that people are lazy and misguided.
Yeah but also that sounds like a nightmare listening experience. Especially if what you enjoy most about music is being exposed to new music.
@@braintrust12 There are plenty of ways of getting exposed to new music via Bandcamp, and there's nothing nightmarish about downloading your music and keeping it on your own device despite what Spotify would have you think.
Timberland is currently working with Suno so I guess it will get a lot harder to detect it… also I know artists have been using this technology prior to this popular technology. Also you look like Kevin McAllister… I like that, just subscribed❤
i'd be reallly curious to see what the payout is like on soundcloud! their fan powered royalties seem super cool!
As a person with ears i heard this video
Bandcamp?
That's what I thought too but bandcamp isn't really a "streaming" site.
Bandcamp?.....does not pay for streaming, is how that sentence ends. If you purchase an album you can stream it, but there is no 'streaming service' on bandcamp and artists are not paid money for streams, only for the purchase of the album. Also, bandcamp got bought out and is owned by Epic Megagames which is killing PC gaming with their predatory business tactics. I took all my music off bandcamp when they got bought.
@ihaveagoddamnplanarthur It kinda is, but not in the same way. I was waiting for some kind of data on BC as well, as I pretty much exclusively get my music from there.. I buy digital albums most of the time and sometimes some merch, like a shirt. It would still be interesting to understand, if and how much artists get when I listen to their material, without buying anything.
Their streaming model is slightly different than whats considered a streaming platform so i guess it would be difficult to compare with others, but i do think Bandcamp and Soundclound are good choices everyone should consider on their de-spotifying journey.
Bandcamp actually doesn't pay at all for streams. Only if and when music is purchased.
"make cool shit, and put it on the internet"
Keep doing that, because it's helping.
Ironically, this is what AI allows more people to do though 😅
it's seldom the video itself that's the cool part. what it's documenting is what counts, and that still takes human creativity and discernment. Generative AI is by-nature derivative, so rarely adds anything of real value.
AI tools are really handy for filling-in the blanks though, but any more than that usually ends with a mediocre result.
so if it's an AI building on my music sample is that also "derivative" and no real value, or not? lets get those braincells firing
@through-faith-alone it's derivative of your work, and of the training.
Sure, there can be some pretty interesting emergent results, but that's not the same as creativity.
@@theafro fair assessment but no, it is the same as creativity. What it isn't the same as, is the same amount of creative work, or creativity.
Is it less creativity? Sure.
Most music is derivative of other music. There's less creativity too.
The difference is in the level of engagement, and use of the tool. In the future, you will see musicians using AI with impunity, as many are already. To create outstanding pieces of music. And you will also see people expressing themselves through AI making music for them. Lastly you will also see mass produced music through AI.
But creativity will never be lost, it's not the tool, it's the user.
You sir are doing the lord’s work, this is awesome 👏
Hey Benn, great video. It would be interesting to see, in your investigation of copyright strikes, if included, there was also a song that was %100 original recorded with live musicians, and see how many strikes that gets. I bet it would not be zero.
The reason most people chose the lower quality music as being the higher quality is most likely due to the fact that the lower quality is their "norm". So many people stream low bitrate music on their day to day, and rarely (if ever) listen to CDs or vinyls, so their mind has made that particular quality the "standard"; and since everyone thinks their way is the best, their mind has set that quality as the best ... You then ask them to compare what their mind has deemed the best to a higher quality, and they're of course going to say that their normal (lower quality) choice is better.
The mind is pretty silly.
And yes, I dork out on neuroscience podcasts all day at work.
The other side of the coin is the systems outputting the music. Air Poops are all the rage and they sound horrible, meanwhile my Sony, Koss, AudioTechnica, and other headphones completely blow those things into the water, with the competition being 4-5x more with the EarBra Max Pros barely keeping paces with headphones that are DECADES old at this point. That also is part of the problem. The smartphone market only had LG, MoonDrop, and Sony the last few years providing the mobile experience that is feasibly achievable outside of buying fiiO or some other extremely niche device for an already less mainstream niche that is HQ audio....... People these days are spoiled with 10s of thousands options, all of them poorly made crap from china with no prior reference to "good" sound. You won't find the majority of people running Focal in their cars or having 5-7.1 surround for their TV unless it comes in a complete "sound system" box.
The thing that gave me hope for humanity regarding music was that Vinyl was making a resurgence before covid and platforms such as bandcamp are readily available these days.
@WhatWillYouFind air pods and old school earphones of the same design are the absolute worst. Horrible quality and sound isolation and you have to put them in your ears too. Go infection go!
Suno's V4 model has a distinct and consistent kinda rattling laser sound artifact that appears all the time. It's very slightly noticeable on V3.5 but very obvious in 4 so it seems their model is converging on it. I'm curious if that's what's being picked up that got your detection rate to 100%.
Suno calls it a "shimmer." I think more prevalent in 3.5 than 4, but just an opinion.
Modern pop artists will probably start adding the "shimmer" to their songs to please the stuffed ears of "pop listeners". A bit like what happened with autotune, which is now the mark of a real pro singer instead of being perceived as cheating.
@@IlBiggo I doubt that, it's a horrible artifact that I can't imagine they're pleased with. Needs to be fixed not leaned into.
@@IlBiggo that's a good one! Very clever.
@@mal2ksc I assumed it was a similar thing to how the early GenAI art looked like an acid trip and full of fractals. Like, the weird resonance was the audio equivalent of those artifacts.
1:50 The Cannabis leaf there for no reason at all to represent the randomness of the video?
I laughed
@@maymayman0same. Cracked me tf up
Are you trying to make sense of things right now?
I did not know what I signed up for when I started watching you from that noise video I watched months ago, now having watched most of your videos and knowing that you chose to stay with UA-cam even with the offer from companies to what I assume is to make content for those platforms is truly king behaviour.
First time coming across your channel and I am impressed by the quality of your work, research and using tech! Nothing better than mixing programming/tech with music production!
Regarding your AI music detector: You seem to emphasize the difference between music that is created with AI being low res (128kbps) while many human songs are available in lossless formaat. Are all samples in the same quality (128kbps)? Otherwise, what you have essentially created is not an AI music detector but a fairly trivial quality discriminator. I think there's a significant difference between the two.
It’s not looking for the quality of the music. It’s looking for artifacts of compression in higher quality audio. Basically, since the vast majority of streaming services provide compressed audio, the AI models think that most music should sound like it’s been compressed, even when it’s generating audio of higher quality. If you can detect this misplaced or ‘wrong’ compression, you can detect AI generated audio.
As a sidenote, you can do the exact same thing with AI generated images, looking for jpeg compression in PNGs, etc.
Some of us can hear it and it's driving us mad
@@manavali92 I understand - that is exactly my point, that the proclaimed 'AI Music detector' is essentially a quality discriminator. It's useful and I appreciate Benn's efforts here, but we should be honest in acknowledging that the use of compression artefacts to detect pieces of AI-generated music is quite trivial and not a robust method in the long run, given this is not that hard to bypass. That's why I argued he should claim that he's using AI to detect low-quality audio or compression artefacts. The fact that it is an effective discriminator for AI music is nice in practice, but not robust - I know this nuance is much less catchy to the average UA-cam scroller than 'Using AI To Detect AI Music', I get it.
Also, how to spot the difference between an MP3 and uncompressed WAV. It's not a matter of detail or clarity, it's artifacts you need to listen for. The hiss of a cymbal retaining it's metallic sheen and not sounding washy. The echo, room noise, and reverb immediately following percussive sounds and how full it sounds and how big the room the instrument is in sounds. They are very subtle things but in those double blind tests you can get >80% accuracy picking the uncompressed audio if you listen for the artifacts.
AI didn't kill music. It was already dying, AI just finished the job.
I've seen some of your videos before, but it wasn't until I saw your "drop" video with Mr. Bill today that I learned you're also The Flashbulb, and I've been listening to your music for years!
Just the conclusion of your video is a freaking great representation of why I follow Venus and you and will continue to do so... you guys are role models for me !
between your video and GamersNexus's commentary today about recent AI news (eg social media companies standing up their own AI bots), I have to say I have not felt more bleak abt the future of AI. humanity trapped paying for endless gray slop created on ever-thinning margins for - being extremely honest - the sole purpose of preserving the egos of the most incompetent, cowardly, and unimaginative corporate management teams in history.
Benn, please, DO something to organize content creators to fight against UA-cam's anti-fair use bias. Fair Use and Copyright are both entitled to the same legal protections, but every time there's a question about it, UA-cam instantly defers to the copyright claimant instead of defending Fair Use.
This has got to be illegal.
Most of the time it isn't even fair use, it's people using samples and loops that they have a license for, but which another song also used.
NO JUST NEON LIGHTS
It isn't, because fair use is a defense you can use in court against alleged copyright infringement. Nobody can say whether or not something is fair use, except a judge. UA-cam occasionally does side with the creator, but in most cases, it isn't up to them to make that decision.
@@DaedalusYoung Fair use is an exception to copyright, not just a defense.
@@jansenart0 "Fair use is a doctrine in United States law that permits limited use of copyrighted material without having to first acquire permission from the copyright holder. Fair use is one of the limitations to copyright intended to balance the interests of copyright holders with the public interest in the wider distribution and use of creative works by allowing as a defense to copyright infringement claims certain limited uses that might otherwise be considered infringement."
A *defense.*
Beatport claim:
"In 2022, Beatport paid an average of $0.10808333 per stream, while its service aimed at open-format DJs, Beatsource, averaged $ 0.17773333 per stream - that’s more than 30 times the industry-wide blended average streaming rates across platforms in the United States was $.0053 per stream."
Checking my own royalty statements I see payments of about £0.02 per stream from Beatport.
Nice! This is cool to know, though I need to verify
who the heck is going to Beatport to stream?
@@blackcatalogue DJs and producers?
@@blackcatalogue Not many, but I have subscribed on and off for a couple of years.
Beatport is a horrible site especially for buying music. They literally have limited the downloads to music you bought unless you're subscribed to their streaming service. They also require you to pay extra if you want to buy the song as lossless like FLAC or WAV which also makes zero sense. In Bandcamp I can literally download the music I bought in any format I want how many times I want with no extra cost.
gotta write some workout pump up jams I guess lol - I guessed Tidal when you said who pays the most and you probably will get it wrong
Hi, Benn!
Awesome video, as always: tons of research, tons of insight only someone with your experience could provide and a fun watch.
I hope you finally found a greyhound with a big blue rubber deek, and I hope you release another Acidwolf album. ❤
Most everyday consumers of digital entertainment don’t care whether a piece is created by AI or a human. They simply want a song that resonates with them. Whichever song delivers that connection wins. People often forget that digital entertainment was originally meant to provide affordable, accessible content for the masses, and AI-generated, targeted songs are just the next step in that evolution. One song = one listener - this is the future. Meanwhile, another group of consumers seeks a deeper bond with an artist, going as far as paying $3,000 for concert tickets to experience a personal connection of a different kind. Digital entertainment is undergoing a seismic shift - one that many fail to see, distracted by fear, anxiety, and anger toward machine learning and AI. We will have no other option but to accept this reality because this is what the masses want. The best thing talented artists can do is to start educating themselves in machine learning or working with ML engineers on a new set of products that will deliver the connection I mentioned in the beginning on a more profound and much larger scale. You could also start touring and playing in clubs, where audiences will soon crave that genuine, real-world connection.
They're too short sighted, closed minded, selfish, arrogant and delusional to see that.
0:37 an artists own website. There I solved it 😅
Except that far fewer people are going to spontaneously discover and listen to that music.
And the artist won't be able to make revenue from advertising, since they're a little website probably won't have any advertisers.
And it's harder to get people to subscribe through a website than getting them to subscribe to a huge music service.
Yes all true man discoverability is more work when you’re directing your own traffic.
But the return is higher, that was kinda my point.
I’m gonna say this again: libraries of many countries already have sort of streaming services and I think that’s who should be taking up the baton.
BTW: my problem with every mp3 versus WAV test I’ve ever seen or read is that they mostly don’t do it with music that I think I’d hear the difference on or indeed that I would want to listen to. I can definitely hear the difference on the live and studio opera CDs I ripped from my own collection about fifteen years ago even though they’re all 320kbps OGG or MP3.
Edit:
I’m not talking about cymbals and other such high frequency content. I think that’s long ago been taken care of. And I admit that I haven’t tried testing my claims for easily nine years. I’m sure a lot has happened in codec development since then
tbh I only noticed a difference between MP3 and WAV when listening to very sonically crowded music, since that is where the compression hits hardest. Since then I'm an enjoyer of lossless.
@ try listening to a single solo flute in a moderately lively room, for instance
I agree. I probably couldn't tell and wouldn't care listening to Michael Jackson, but play me some chamber music, some acoustic jazz, or even the high definition Dark Side of the Moon and I can tell you. I also have pretty high quality audio equipment because I DO notice.
I mostly only listen to metal, and its pretty easy to spot a low bitrate song, the cymbals and hi-hat are a dead giveaways if the bitrate is very low. Just sounds awfully crushed and hissing.
I say that I can tell but even Ben admitted that at 256kbps he lost that ability. I would love to see a video of all types of people from pros to musicians to younger and older to audiophile to average person try to define Flac/Wav vs mp3/opus/aac/oogVorbis.
For me, specifically hihats and cymbals tend to lose that shimmery sound. That is normally the giveaway.
Could it be that older folks with their hearing slowly getting more and more low passy might have a harder time identifying format?
If that is true there would likely be a peak where someone's ear is trained more and they still have not had that decline start. Something like a sports player with the experience and joints all still intact...
It was great to hear on your views on lossless audio formats. Personally, i do notice a difference with formats sometimes...it definitely depends on how the track was mastered. If the track was designed with lossless formats and audiophile grade gear in mind, thats where you hear the difference. Once you get used to a good setup, you do start to notice the drawbacks with mp3 files vs lossless.
You are, without a doubt, one of the most interesting and diverse UA-camrs I know of. You seem highly intelligent, and you're able to break down complex topics and present them in a way that makes them understandable and entertaining. Keep going, please!
The question of which service pays artists the most per stream, is a nonsense distraction.
It ignores that the revenue of the artist has more to do with which service will spontaneously increase their listenership,
more than the hard number that is paid.
this. i dont give a rats ass about the per-stream revenue. All i care about is the algorithm.
@@braintrust12 Unfortunately for you, the algorithm will prioritize the streaming services' own mass-produced background music. Or maybe they will just add a couple of millions extra listens on their friend's tracks, as Jay-Z did for Tidal.
@@Mnnvint the algorithm cant make people voluntarily listen to ai music. because ai music sucks. if it was good, then sure the algorithm could get people to listen to it. but it sucks so nobody is listening to it
@@braintrust12 Yes, the algorithm can make people listen to ai music, because a lot of people just put a spotify radio on repeat, and ai music can be "good" in the sense that it sounds like the other music it's presented along with if you're not actively listening.
@@Mnnvint i just don't see it. i've never heard an ai song that is good enough to blend in with other professional artists' work. i say "professional" because i've heard stuff from amateur/hobby producers/writers that sounds so bad that I literally think to myself "wow... this sounds like ai made it." in fact, that's the only purpose ai music serves atm: a benchmark by which "terribleness" can be measured.
What about people who write their own lyrics, but use Ai to generate the music and vocals?
Benn...I love you
Love you too beebs
i love that you literally left your video in log colors. looks actually nice.
I kinda want to get into the visual art version of what you're doing as an film animation student. I also have some sound design under my belt and am learning about sound/music: it's a lot of work! People deserved to get paid for their work! Props for all the work you're doing and sharing. :)
I was once in a party with a paid employee of a local bank, which is also a huge IT company branching out into different things, AI included. She worked in their music AI department, she said she was in charge of training data. When I asked her what's the process for getting the music, she said "Torrents, lol".
I gave her the coldest shoulder I could. But I doubt there's going to be a comeuppance for this, unfortunately.
0:30 It’s gonna be Deezer, isn’t it?
Music : " our brains are collecting pressure waves and then passing them through latent inhibition..."
NEON LIGHTS
You're awesome, research like this is incredibly eye opening, and concerning.Thank you for doing work like this.
I love this man, thank you :) ❤