The Vergecast - Can AI make a hit song?

Episode Date: September 18, 2023

In episode two of The Vergecast's AI mini series, David Pierce sits down with Switched on Pop's Charlie Harding and music producer Ian Kimmel to share how they made an entire song from scratch using a bunch of AI tools. Later, Nilay Patel joins the discussion to talk about the future of AI in the music industry. Email us at [e-mail] or call us at 866-VERGE11, we'd love to hear from you. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Support for the show comes from Retool. Too many companies run critical operations on duct taped spreadsheets, Slack workflows, and whatever else they could cobble together. Not because they want to, but because building internal tools means weeks of waiting on someone else's backlog. That's where Retool comes in. Build custom internal tools just by describing what you need. Prompts something like,

Build Me a Revenue Dashboard on our Salesforce data. And Retool actually builds it on your company's data, in your cloud with enterprise security built in. Go to retool.com slash Verchcast. We all need to retool how we build software. What's up, y'all. I'm Skylar Diggins, seven-time WMBA All-Star, Olympic gold medalist, and mom. And I'm Cassidy Hubbard, host and reporter for nearly 20 years, covering the biggest names and stories in sports and mom. And this is Am Mom, a community for athletes, game changers, and moms of all kinds.

Tap in with us. Welcome to the Vergecast, the flagship podcast of procedurally generated four on the floor. I'm your friend David Pierce, and this is the second episode in our three-part series about AI. What we're trying to do in these episodes is not just talk about AI as a theory or fast-forward a dozen or a hundred years to when AI takes over the world and kills us all, but actually look at what AI is really doing in our lives right now. This week, we're going to talk about how AI is changing music. And we actually got started down this particular rabbit hole while we were making last week's episode about cloning your voice. Because in the making of that episode, we discovered this tool called voicify.AI that claims to be able to train a model to sound not like your speaking voice, but your singing voice.

As far as I can tell, Voiceify isn't really a production tool so much as it is sort of a silly tool to let you make songs using voices like, I don't know, SpongeBob Squarepants. or Michael Jackson's voice. But you can also train it with your own. But you can also train it with your own. So obviously I had to test that. So I got out my mic, put on some headphones, and I spent about an hour singing a few of my favorite songs into a recorder. I'm not going to play those recordings for reasons of my own embarrassment,

but also because trust me, literally no one needs to hear that. But the thing is you're kind of about to hear my voice anyway, or at least what voicify. AI thinks I sound like. The process is kind of weird. You upload a bunch of stuff, it processes your voice, and then you just plug in a YouTube URL. It tries to extract the vocals, then sort of paste yours back in, and then rewrite the song using your voice.

It's wonky, but it's super fun and pretty quick. So, without further ado, please welcome to the stage. A.I. David Pierce singing, I want to hold your hand by the Beatles. It's actually not bad. Like, I cannot stand the sound of my own singing voice, but that does kind of sound like me, I think, at least until you get a little further into the song, and then it all just completely falls apart. Okay, I could do this forever. We made a lot of songs, but I'll only do one more here.

This is Suspicious Minds by Elvis Presley by AI David Pierce. We're calling a trap. I can't walk out because I love is too much, baby. Oh, Lord. Okay, let's just move on before this goes any further off the rails, or everyone stops listening because of my voice. Let's bring in a couple of actual musicians to talk about how, AI is changing the way we make music. My guests today are Charlie Harding, a songwriter and the co-host of

the wonderful podcast Switched on Pop, who has been on this show before. He is responsible for Lasercong. It's a laser bomb. It's a bomb. They're all lasers. And Ian Kimmel, a producer and songwriter who has worked with everyone from BTS to Mary Jay Blige to Netflix and the NBA. Ian also runs a company called Biscuit Head Collective, which works with artists of all sizes to make songs. So you could say AI is very important to both of these people and their livelihood going forward. And I want to know how they're thinking about it right now and how it fits in. Charlie Harding, welcome back to the Vergecast. Thank you for having me. This is just delightful having you back. It always gets weird when

you're here, which is my favorite thing about having you here. It's going to get weird today. Ian Kibble, welcome to the show. Thank you. Thank you so much for having me as well. Okay, so I want to do this episode in two parts. Sure. Talking about AI is very complicated because you end up having to talk about about like the cool things it can be and then the terrifying things about it and then we're all going to die and it's illegal all at the same time so we're just going to split those things apart so for the first half of the show i want to talk like super practical how this stuff actually works what you can do with it you're both professional musicians i want to talk about like how this actually

works like in the world and then we're going to take a break we're going to collect ourselves and then we're going to talk about our feelings does that sound good you guys good at that okay so i gave you guys some homework and we're going to get to it in a minute and it's very exciting. But I think the first place I want to start is like if you're a working musician today in September of 2023, are you thinking about AI in your day to day at all right now? I am as a music reporter. Sure. But in terms of making music, I think we probably share this. The only tool I've really used to assist in any part of the creative process has been chat GPT, but even that is quite limited. Interesting. That's not at all what I would have guessed. Oh, yeah. So how are you using it right now? Like,

where would chat GPT fit in? You can use it as a rhyming dictionary. Ooh, okay. But it hasn't come into my life very much. I mean, I wanted to bring Ian into this conversation because I feel like he kind of has, I think he has the most skin in the game. Sure. Because on top of being a record producer, he's got, you know, platinum records, Grammy nominations. He also runs a business where he will help you take your indie track and make it a like top 40 professional style song. So if there's anyone who probably cares about the world of AI replacing musical things, like, I mean, he's directly invested in it.

So I'm curious. How has AI been in your life? I mean, the number one thing to think about, at least when I'm jumping into it, is generally, you know, we go back a couple decades and we're looking at people like going to tape and that kind of thing. And then we have things like VSTs, the plugins replacing all that. So now we're getting into this new era of AI and like all these new technologies coming out that the number one thing I'm paying attention to is just trying to be current. I'm young. I'm in my 20s. But at the same time with all this technology we have to do, like it has to.

We always have to keep up with the times, otherwise we're going to get left behind. But there is so many amazing functions going on inside of it. And me being a music professional who is a musician that can play multiple instruments and program drums and all that kind of stuff, it doesn't come into my world as often as like a new music producer would. Okay. Just because there's so many easy tools. So you think if you were starting fresh right now trying to get in the game, you think you would be maybe more invested in figuring out some of the AI stuff? Definitely.

Yeah. If I didn't know how to program drums, there's, you know, you. you can just pull up this one VST that allows you to program drums with a hit of a button. It automatically reads your tempo. You can set the vibe that you want. And then you've got drums. And then you click the button again. And then you have a variation of that.

And it just, there's so many tools that it's creating to really save time. And also in the learning process, it's a bigger jump. It's a lot quicker. So it's allowing new music producers to be able to jump on way quicker. But that being said, the thing I only really use in my professional life would be chat GPT. For the same reason, rhyming dictionaries, all that stuff? Not as much for me. Sometimes lyric editing to see if I can pull something better out of it, not as often,

but the most I'm actually using it for is writing well-converting business emails. That's the top thing I'm using chat for. That's your hit album, Well-Converting Business emails. Yeah, exactly. Sounds like a great vapor wave album. Let's make that. But yeah, that's the top thing I'm using it for as this time. It's the sidebar, the music industry,

is the business of the music industry, but it's something that's very important for somebody like me who's trying to afford to live in New York, you know? Yeah, for sure. So is there anything that feels like you would love to offload part of the process to AI where I feel like, you know, everybody in who does sort of normal desk jobs is like loves AI because it can write their emails, right? And that feels like the kind of thing that it's like, this isn't really my job, but it's a thing I have to do as part of my job.

So like terrific if I can offload that amazing. I mean, we were talking about this just yesterday. I'm very jealous of Ian because he has a team of people that work with him to help him do all the prep work, which is really boring. So prepping a mix session takes forever. You have to cut out all the silences. You have to get everything in the right order. You have to color the tracks. Tuning vocals, aligning things.

A lot of things do have creative choices that happen in them. But there can be just dozens of hours a week of prep work that usually is handled by an assistant or an engineer. And I don't finish songs often because I'm like, I don't want to spend my entire day doing all this prep work, which not only slows me down, but it also, it drags me down as well. Like, it puts me in a creative hole where what I need to do is just, like, get right into my creative state, but I have to do that five hours into staring at my computer. So there's a lot of things that I would love for it to automate. And at this point, it's not replacing anything I'm doing or my assistants are doing. But there is technology now coming out that is helping with it.

but I think we're just on the verge of that. I saw one last week that, so we talk about the stems of a song, all the different instruments and vocals and everything. And if someone, you know, a lot of times we'll get it from the producer or the artist and nothing will be labeled. So we have to go through and listen to every single one.

And now there's a new technology where you can enter in your stems and it can name exactly what they are and you can choose how they're named and everything like that. So at this point, there is new technology coming up that can really save us a lot of time, save my assistants in particular a lot of time.

but we're just on the verge of that, which is personally really exciting for me. And it's important because, like, if you were to mix the artist Jacob Collier, was a Grammy winning artist, what is a small session of his? Like 400, 500 tracks? Yeah, definitely.

Good Ford. Right? So if those haven't been labeled, you've got 400 to 500 tracks. Apple literally expanded the number of tracks that their DAW logic can have because he was feeling like he was hamstrung by the number of tracks they could have.

So you can have these huge sessions. If things are unlabeled, I mean, it could take an entire day just to get things set up properly so you can make so much time. Yeah. That's a really good example. I mean, it is kind of the like the work about the work. And that's so much of AI. Like we talk about this even with like what Google and Microsoft are doing where it's like you can't do your actual sort of meaningful work better.

But you can turn an Excel spreadsheet into a PowerPoint deck more quickly. And that's where we are with AI in a lot of places, I think. And that's fine and good. But yeah, I just feel like there's a there's a leap away before it starts. being more than that. Like, do you feel like you're in a moment of existential business questions about, like, how do I do this job? We're going to get to much more of that later.

For sure. Preview it for us now. The toes in the water. Yeah. No, it's definitely, I am not worried at all, nor should anybody in my company be worried about, like, you know, it's just minuscule tasks at this point. There is, you know, there's things rolling through my head, like, okay, we have the STEM technology to read the stems.

So that means, you know, AI can't hear critically, but it can read waves and all that kind of stuff to really understand what's going on. So it can tell what different elements are. You can reference things that sound good, and then it could like mix your song in a way that sounds good. And, you know, that is not at a point we are at right now that technology hasn't been made. But I see it in the future for possibly being able to automate some of the bigger jobs that we have. But just the data set that you would need in order to create that would be incredible. You know? Totally.

Yeah. Music is, music creation, though, a big business is often a lot of, a lot of the companies that serve the music professionals, they're small businesses. I mean, some of them are multimillion dollar businesses. But, you know, if you look at like the synthesizer companies, they're like a dozen people. They're using really old technology from the 70s often. There are a handful of, you know, certainly music technology companies that are working in AI, but they don't have the resources of Google, right? In fact, some of the tools we've used for fun have been created by, Google as sort of like side throwaway projects.

And so one of the big problems, I think, in servicing music professionals is large good data sets that might exist for, you know, helping us make PowerPoints because how many PowerPoints exist. But getting access to good data, I mean, I remember the Verge had an amazing article written by a, you know, former journalist here, Danny Deal, who talked about the metadata problem within music. And just like, people don't submit the metadata to the streaming services for you to then know how to get paid. So there's just like the data problem is already such a significant problem

throughout the music industry more broadly, making tools that are built off of that training data for our subset, fairly niche industry. I'm, you know, skeptical and expect the best tools probably being things that are appropriated from the larger tech companies. Got it. Okay. All right. So the super primitive nature of all this makes me very excited for what's about to happen, which is that I basically told you I wanted to talk about some of the specific tools that are out there and how they're using them. And the two of you went like several orders of magnitude beyond that. And you just made AI music is essentially my understanding of what happened here. We made music using only AI tools. Okay. I want to walk through this whole process. I want to hear the thing that you made. I have

not heard any of this. I'm very excited. But before you get into it, how was this process? You did this with like very little notice or time. Was this like, was this fun? Was this awful? Did you make the worst thing you've ever made in your lives and you hate it forever? Like, how did it feel? I feel like, Ian, you make stuff just every single day. How is this different for you? I'm curious. I'm so happy I was doing it with him, first of all. A lot of other people, this would have been so frustrating. But I felt like I had a broken leg the whole time. It slowed down the process. And it's not just like figuring out these tools and how to use them. It's how to like, we had to restart and manipulate audio in a way after using the AI to actually get it to fit in correctly, which was so time-consuming when it's

like one of the tools we used was just to create a simple base, which was just a sign wave. And I could have pulled up a plugin and done it in 30 seconds. But in order to try and create a sign wave using some of these things, it took us 10 minutes, you know. So it was just a slow process just going through the entire thing. And, you know, it's really fun getting starter ideas and little things that I personally wouldn't think of. and pulling those in, but using it for the entire process was very inhibiting. Yeah, I believe that. Yeah, I mean, learning any new tool is always going to slow you down, like, with any studio

tool as well. Like, if you get new equipment in the studio, you've got to kind of spend an hour learning it. So they're definitely, we were held back by using new things while trying to be creative at the same time. That said, most of the material that we put together would have been a lot easier if we could have just used the existing tool sets that we already have. So it was a fun exercise, a set of constraints, and a fun excuse to hang out and make a song together. Love it.

So let's do it. Walk me through. Like, you sit down to do this. Where do you even, where do you start? Well, I had done a little bit of research the night before to make sure. And, you know, I cover music. And so I had a list of things. I was curious about exploring.

And we thought that we should probably begin with an idea, right? Like, typically a good song is going to have some kind of hook, some kind of core idea that's going to guide it so that we know. which direction to go in. And we figured we could use chat GPT, which is the, you know, that would be the top of the funnel. Well, everyone knows it. Let's see if we can use that for some sort of creative purpose. And so I came into the room and I had been talking about, I've just started a teaching position, which is really fun for me. I'm teaching a music school. And I actually didn't finish my music major. So there's kind of this like imposter syndrome that I have of going into class every day and teaching something where I kind of dropped that major. And so I thought, well, let's go with that idea. And then I think you went a little further. Like, how about let's let's take it to like just the feeling of like not belonging. And so you made an amazing, you have a way of prompting chat GPT in. that is sophisticated and results in some lyrics. I don't find it sophisticated, but thank you. Yeah, so essentially he had the imposter syndrome idea, and I was thinking like, okay, what can we make this more standard song core,

lyrical content-wise? And so what I thought of was like, okay, I have imposter syndrome, but imposter syndrome with somebody else, it would be really easy to jump into a relationship. So I prompted a little bit of relationship, and then I also prompted it with I have imposter syndrome, being in this musical or in any world with AI also there. I feel like it's better than me.

I feel like it can do better than me. So a lot of the nouns and things like that that were verbs and everything were like electrical techies, stuff like that, which is really interesting. And basically writes like a multi-paragraph long prompt that is just like you have this way of just kind of like word vomiting. Stream of consciousness. Like maybe that will prompt interesting and creative things. And I think we came to the idea that we would like this song to be maybe from the perspective of an AI that has impossible. Stendra I'm trying to exist in the world. Ian has this amazing studio in Midtown that has a, they're on the eighth floor with this

giant window that overlooks the Empire State Building. So we had to make it also set in New York. Sure. So we sent that prompt in and we got back garbage. He used chat GPT 3.5. I used four. I used four. I think four is supposed to be the more creative.

And so this is what we got from our just first verse after this big prompt. Woke up today in a world so bright, but shadows form. beneath the city lights, you're soaring high while I remain tied to the ground by invisible chains. Starting a song with Woke Up Today has got to be the most cliche way of beginning a song. It's like an off, off, off Broadway show that we're making here. And so the real challenge here is it does write, yeah, the most cliche kind of lyrics. Very obvious end rhymes, bright lights, remain chains.

That's kind of nice. It's actually made a slant rhyme. Sure. It's really bad at like multisyllabic rhymes, which is what makes, you know, a great verse, strange rhyme schemes and things like that. We were also upset that it was kind of vague. Good songs are going to have some specifics. Like you want the red scarf and a Taylor Swift song that's going to guide you through the entire narrative. And so we tried a different approach.

I borrowed this songwriting tool from Jeff Tweedy of Wilco. And he's got this idea where you want to hack your brain and put together unlikely pairings of words are often where you find interesting new meaning. So I said, give me, this is one of his exercises, give me 10 verbs that are about imposter syndrome, and then give me 10 nouns that are objects in New York City. And it started giving me more interesting poetic things, including intruding skyscrapers, shattering street lamps, looting newsstands, tampering taxis. I said, okay, well, take some of those and try to write some lyrics. So it gave me this verse, intruding skyscrapers, piercing the night.

in their looming shadows, we tried to find the light, sneaking subways, silent cries they echo, were in different lanes waiting for the metro. Better. Better. We liked metro, although clearly nobody in New York City calls the subway the metro. Sneaking subways kind of works for me. I like sneaking subways. And then I was like, wait, we need more specific things.

So I then prompted it further. I was like, but can you give me some specific places in New York City? Now we go down the rabbit hole. And it updated it to intruding skyscrapers. The Empire's Might by the Flat Iron Building, we first saw the light sneaking subways beneath the Central Park by the Bethesda Terrace. We lost our spark. And you immediately start singing. Do you remember?

No, no. The Alicia Keys saw. Oh, yeah. In New York. It was just so, like, incredibly specific or, like, you know, I love an actual place, something tangible. Sure. You know, like, I put the word Bowery in so many of my songs. That's ridiculous.

But not every line. Yeah. It's, yeah. So we kept prompting and kept prompting and trying to get better material out of it, and it just wasn't working. And so we decided to go with sort of the, I think what was the second prompt, the one that used that sort of noun verb exercise. Sure. And Ian is really good at reducing bad lyrics that's been submitted to him and make them, you know, more singable. So we knew we had something he could work with.

I think my favorite AI tool, actually, that we used the entire time. I'm assuming it's AI? I don't know. was he took a picture of my computer screen with the lyrics on it, and Apple automatically can identify the text, and he could copy and paste the text from the picture into an Apple note and edit it. I think most of... I love that that's the best AI.

That was the best thing that we used. That is so telling. And I think the tools that I had the most fun with are the things that often market their AIness the least. So I don't know if there is some sort of machine learning in that tool. Yeah, no, that counts. That's AI. Yeah, we're good. Great, great.

Great. All right, we got an idea. We've got some rough lyrics. We need a vibe. And so we went hunting for sounds. And the first place we went was a website called Suno. Suno is kind of like the mid-journey for music. They have a Discord server where you can enter a prompt and it's going to spit out rather than a picture like Mid-Journey. It's going to spit out music. And it's, I believe, fully synthesized music, including lyrics and vocals. So can I play you what we got? Please. Okay. We're just going to do a smattering the sounds for now. We entered the first verse that intruding skyscrapers verse, and this is what gave us. Intruding skyscrapers, piercing the night. Rolling shadows, we try to find the light. I was like, okay, I like the Nokia ringtone. We're going to use that.

And we've got some vocals. So all of that was synthesized. You said it text, and it came back with that. That's right. Also, this is the full audio quality, like, to be told to. too. This isn't being degraded at all through the listening process here. This is the full audio quality. It's basically Nokia ringtone quality. Yeah, yeah, exactly. Okay. And then actually just like Mid Journey, it will give you alternatives. So it gave us this other track as well. And this one we thought

was kind of trash. But it has this. Yeah. I'm getting like bad country song vibes. Yeah. But we like this little sort of pianoy synth thing. Now we have the problem that we want these individual pieces of material to play with. So we go to a website called AudioShake.AI, and they will stem your pre-existing music for you when you're all separated. Audio Shake will take that track, and it gives us the vocals, the little piano riff we liked from the first one, and some of like other stuff. And then you can listen to them individually. So this seems like the kind of thing that I could see being useful for lots of things in the world, but not really in the music making process. Like, are you often pulling stems out of existing things when you're making music. Well, I mean, generally,

that's against copyright. Well, yeah, exactly. People use it for remixing. So especially when, when, like, it's a really old song. And so you do have the rights for it. Big name artists are actually using this tool to be able to get the stems to be able to remix. Right. Yeah, you also see it all over TikTok and stuff. They're using it to, like, isolate the vocals and stuff like that. Yeah, yeah, right. Okay, so isolating vocals. We now have isolated vocals. Okay. Right. And we have a little fun riff. this little Synthy thing. I do like that. My Nokia ringtone is nice.

Okay. So that's the first sound we assemble. We've got Suno. We also pulled... Piano riff. That thing from the Bad Country song. We got that as well. All right.

So we got those sounds. I then just went to a ton of other... We will make music for you AI sites, thinking maybe I can just keep stemming stuff. And a lot of them are really bad. The quality of music that we were getting back would not fit like top 40 pop kind of vibe. There are much more like background advertisement,

sort of mood music, but we figured maybe we could, you know, get inspired by one. And so we went to a place called Soundful and heard this music, which I think sounds like a pharma ad. Side effects include dry mouth. Yeah, yeah. Exactly. And we're like, okay, so we took this to Audio Shake and we got the piano.

Okay. The quality is not great. But this guy's a wizard. We knew he could do something with it when we play with it. We got a few more sounds. We went to Splice is one of the biggest libraries of samples. And they have a new tool that will automatically pair different samples together for you as sort of like a starter. Okay.

And we went to their beta AI tool and they made this sound for us. These are four different samples that Splice just kind of randomly put together. And we thought sounded kind of like a really bad phone call again. Yeah. Like you're on waiting. But we liked the kick drum. So we got the kick drum. Okay.

Full drums, yeah. Snared too. You had a sample of your own. Ian has a professional producer has a bunch of starter sounds that he can bring to any session and just like get things going creatively. But we had to figure out how to process something he'd already made. made through AI. Sure. And so there was a tool by a company called Output called Arcade that will automatically take

your sample and play it back to you with some kind of AI tool that will make some creative choices with that sample. It's essentially the listening and chopping. Arcade also does a creative function where you can set the key and it looks at the tempo. Okay. And then it will give you sounds that could work. It's not using any kind of learning technology in this case, but they have a new portion where you can input a sample.

It learns that sample figures out where to chop it and then plays it back to you. you in creative chops. I see. Okay. So here's the chops that it gave us. All right. Now we have a selection of sounds.

Basically, we have lyrics. We have collected our paints. We got to now put something together. All in different keys. All in different tempos. I think that whole process took us like two and a half hours of getting all that material and just like getting something to work with. We have a lot of piano tracks.

A lot of piano tracks. A lot of keeping track. Yeah. I'm feeling this is going to be a very moody song. We were thinking moody. Because we actually made a, why didn't say we did make a little reference playlist. of artists that maybe could inspire the vibe. We definitely wanted moody piano was on there.

We wanted a sort of mid-tempo, happy, sad, walking through New York and the rain kind of thing at night. But it's like neon all around. All right. So that's the vibe. And with all of that prompt, we probably could have written a song or two in the amount of time that it took us to collect these sounds, which are very low-fi and need a lot of processes. It makes sound good. So, Ian takes first the piano and gives us... this riff, the piano from Suno.

The piano from Soundful, actually. Which, you know, sounds a whole lot better than... Oof, yeah. That's where we started, yeah. That's what we start. And then I think we had these, like, interesting... We had those, like, sparkly sound effects. There's no idea, yeah.

We piled them together. And then we had that kick drum. It took us probably a half hour to process the sounds to actually sound better than they initially did. we thought, okay, that's sort of like a pre-chorus chorus. Now let's use some of our other sounds. Let's make a verse. And one of the issues is we didn't have a bass,

and we were trying to limit ourselves to only the sounds that we had our disposal. So we took this tiny little sample from Suno again, Suno, the sort of mid-jurnary, like app. That. Just that. That. And I turned that into a bass sound. That's like a space base.

Space base. By like staring and fiddling. Basically, I put that into a sampler. I process it. I do a bunch of different things to it to give us some kind of bass something. And I think I use the same sample to make this as well. It just took so long to get from that one to another one. We just had to run it through like three plug-ins.

And yeah. Okay. So, and we have one more sound, which was this. I can't even remember which one this came from. That was for my loop. This is your loop. This is your loop. This is the arcade loop.

Okay. And so kind of all those things together, give us a verse vibe. And then I think there's just other little chirpy sound. As the piano. And the sounds are like, they've got a lot of texture, we'll say. I think they sound kind of like the Nokia ringtone is probably the best metaphor. But for me, it's like those really bad MP3s that you definitely didn't download off of Kazah or Napster, you know, 20 years ago.

That's kind of the quality that they came in. Yeah. Yeah, there's that one that kind of makes you want to like peel your headphone off just to make sure it's like not you didn't blow your speakers. Yep. So the next thing is, okay, we have, now we have a vibe, we have some lyrics, we've got to put together some kind of, we'll call top line. We need a melody that's going to go on top. This is where I think we gave up on trying to use an AI to write a melody. I was just like, Ian, get in the booth. Sure, I just started riffing and see what it works. He puts on some auto tune. Here's what he gives us. Intruding skyscrapers Pairs through in the night

looming shadows under the city lights Not bad he kind of has massage the lyrics I will say that I think intruding skyscrapers is a terrible opener to a song It's very wordy

Intruding skyscrapers There's a lot of syllables to say At the very beginning Doesn't roll off easily So we build a whole bunch of melodic ideas pair them with these lyrics. Ian puts down a whole set of vocals to the entire song. The more interesting thing, though, is we can't use his voice.

We have to use someone else's voice. So we send this through voice swap.aI. If you want to release a track with them, they will actually allow you to license their professional singers. This is one of the few tools that has actually made pretty high-quality tools specifically for musicians. So they're actually generating models with specific musicians, and then you can plug your own stuff in and get their generated stuff out.

If you license it for a couple of different. $100 basically. So this is for journalistic demo purposes only. Of course. We're not releasing this song. Do we even name for the song, by the way? Have we titled it yet at this point in the process? I don't belong probably.

Yeah. You haven't heard the chorus yet. Oh, you're right. I apologize. All right. I don't belong parentheses intruding skyscrapers. That's it. Perfect.

All right. So we send his voice through the voice model. Intruding skyscrapers. Kirst through the night. Looming shadows. Under the city lights. Now, it sounds very process and affected because we intentionally took his auto-tuned vocal and sent it through the model.

We want this to sound like it might be an AI singing to us. I was going to say it sounds more robotic in a good way. Yeah, we didn't want it to sound natural. That was the goal. Probably voice swap is supposedly the best at doing this, but we tried to hack their thing to give us bad quality material back. There were a few other small tools we used at this point. I mentioned that Google has a handful of projects, one called Magenta,

where they will edit MIDI data. That's basically, you know, the programming language that musicians use to input music into their software. And we had made this little rhythm on our own, just that. And what Magenta would do is it will give you variations on your rhythm. So it gave us this. Okay. I mean, one of us could have done this. But we just figured, all right, there's one natural variation.

It's interesting how many of these tools are doing kind of roughly the same thing where you, it's like, I mean, they're basically all chat GPT. right like you say like here's a thing give me other versions of this thing and it's able to do that with sort of some level of competence but there's so many moments in this process it seems like where you've said like okay we need to put all these pieces together or we need to

start from scratch and get somewhere and these are just they have nothing yeah I mean because like I just need one little percussion idea and I honestly don't there's probably some tool somewhere but if you ask me can you like go interpolate and make me an interesting little top percussion moment like it would have been faster to program but anyways

I used magenta to make this little it's intentionally buried in the back of the track and there's a artificial tape plug-in to sound like a tape machine that's trained off of a neural net that we put in here to process made by baby audio. Finally, I'm going to play the whole thing in a second. Love it.

We put the whole song together and there's a company called Isotope that has a mix and master assistant tool. Their mixing tool is called Neutron and that will go on every single track and you can basically say, hey, mix this track for me. And at the end, on the full master track, you put ozone.

And same thing. Rather than mixing and mastering the song ourselves, we just hit the master assist with a license that ozone gave, that isotope gave us to make this song. And isotope specifically is pretty widely used, right? I don't know how much the AI stuff is, but the isotope thing is pretty well-known. And this is a great example of like the AI marketing around isotope and ozone. It's just like a tiny little slice of what that's being used for. I think they're mostly chasing after the people who are doing the top of the top work and know this sounds so well and just want to make a really big help tool in order to jump on.

They don't want to necessarily take someone from like zero percent to 10 percent. For sure. Yeah. That makes sense. So anyway, do we hear the track? Yeah. Here's what we got. This is, I don't belong, parentheses, intruding skyscrapers.

Intruding skyscrapers. Here's through the night. Looming shadows under the city lights. sneaking through the subway styling We're in different lanes Waiting for the Metro We're going to take a quick break And then we're going to come back

And talk more about this But we're going to let this play us out Support for the show comes from Framer Framer Framer is an enterprise-grade No-Code website builder used by teams at companies like Perplexity and Murrow

to move faster With real-time collaboration And a robust CMS With everything you need for great SEO Not to mention Advanced Analytics That include integrated A-B testing, your designers and marketers are empowered to build and maximize your dot com from day one.

So whether you want to launch a new site, test a few landing pages, or migrate your full.com, Framer has programs for startups, scale-ups, and large enterprises to make going from idea to live site as easy and fast as possible. Learn how you can get more out of your dot com from a Framer specialist or get started building for free today at framer.com slash verge for 30% off a Framer pro annual plan. That's framer.com slash verge for 30% off. Framer.com slash verge.

Rules and restrictions may apply. Support for the show comes from LinkedIn. If you're a small business owner, you know that every hire counts, but time and resources are limited. Finding, connecting with, and screening the right candidate. takes up valuable time you could be giving to your customers. That's where LinkedIn Hiring Pro comes in.

It's built to be your hiring partner, helping you find the right candidates faster. That way you can hire with confidence without turning it into another full-time job. Hiring Pro streamlines the entire process from drafting your job to shortlisting candidates and conducting AI-powered interviews for initial screenings. It's updated conversational interface

lets you describe what you need in plain language. Nearly 60% of hirers find a candidate to interview within a week. With Hiring Pro, you spend less time searching and more time connecting with the right talent. And instead of getting buried in resumes, you get a focus shortlist that actually moves your hiring forward. Join the 2.7 million small businesses using LinkedIn to hire. Get started by posting your job for free at LinkedIn.com slash track. Terms and conditions apply. All right, we're back.

We brought an Eli Patel. Hi, Eli. Hey, what's up? It's time to talk about our feelings. But first, I want to know, you were sitting in the control room listening to the song. What do you think? Does it a song? Do we have a banger on our hands?

I think someone needs to go in and move those drums into the pocket. They're early. The snare was early. It's like, I can't hear anything beyond that. And that's just me. You should have so many more critiques. I'm confident there are more critiques to be bad. It's really hard for like my musical representation on this show to be.

It's very difficult for me. So do you listen to that and think like, is there anything in that that's interesting to you as a musician? The more I've done this, the more I produce music for like so many different people, the more I've realized that I do make bad music and that's okay and that's going to go right into the trash can. And that is just fine. I'll make a better one tomorrow. And, you know, it was really fun using all of these tools, but it would take me more time to go into this and try and save stuff from it,

then it would just be to make a new song. Yeah, yeah. That's fair. My old feeling about AI in general is it is a canon of C plus content. Yeah. And if you point a canon of C plus content at any business model, something happens. And that song, like, it's a, that song is like a canon blast of C plus content.

I think the biggest problem was that we used that piano, which came from what sounded like a pharma ad. And so we started from a terrible, source material. I remember when we were when you were working with it, we're like, we're not going in the right direction. What are we going to do? And I was like, we got to make the, the verse has got to be minor because it's too happy. And it just like, we ended up making like slightly better remixed pharma music. So I don't know. I mean, look, have you listened to the radio lately? It's not, not what the kids are. Okay. So the piece of this that I think is left to talk about

is the like what this means. Like the capital F feelings about all of this. Right. Right. And I think, Neely, we've spent a lot of time in the show talking about, like, what AI does to these industries and how we use them. But I'm especially curious. And we talked a little bit about this at the beginning. But like, as people who make music professionally, we're in this moment where I think most of these tools are not very good. Many of these tools are useless, as we were talking about at the beginning. But like, there are a lot of people who would tell you we are on the road to this stuff becoming a part of life. Right. And I guess the first thing I wonder is like the story of technology helping us make music goes back forever, right? And there have been these sort of moral panics about autotune and the idea that you can make drums on your computer instead of having to play drums and everybody gets up in arms about that. Like, is AI different from those things or is it just kind of another technology on the spectrum of how we make music, which is a forever changing thing? I don't think this is any bigger of a step than we've had in the past. Just because a lot of the AI technology that we have is just doing little steps if you really look into it. So we have like splice that has every loop under the. sun that I can just drop into my song and it can or cannot work, but I instantly have a great

quality, let's say drum recording. And then you have programs like Drum Monkey, which uses AI. I'm not actually sure how, because I'm pretty sure it just reads the BPM of the song, which is incredibly easy to do, and then you choose the genre you want, and then you hit generate. And it gives you a drum sample pattern. And you can hit generate again, it'll give you different drum sample pattern. And to me, that's kind of no different than going on Splice and going through those sounds and picking one from there. It's automatically syncing up right away so you can hear it quicker. But there isn't a giant step anywhere in music that I have seen at least. I think a great analogy would be like the release of drum machines in the 1980s.

Originally these things were made to be like demo devices, maybe replace a drummer, and a lot of them were completely not. They were failures. The most important sound in contemporary popular music is probably Roland's TR808, the drum machine. it was a failure, ended up in pawn shops, and early hip-hop producers start thinking, well, that's a cool sound. If I misappropriated and use in these different ways and I mix it in an interesting way, this thing which sounds kind of like a toy actually sounds amazing. And it took decades for these rejected devices then to become the mainstream sound.

I think what will be most interesting are the weird sounds that you can get out of these tools, kids in their basements, figuring out how to misuse them, at least at the state we are today to create new interesting creative material. But this was mostly a really challenging exercise, with the exception of isotope's tools were very usable, very useful. I mean, we use those everyday professionally anyways. Apple's iPhone was very useful. And chat GPT was mildly useful. You know, you brought up Autotune. And the panic over Autotune, it's important to just remember when Autotune first hit is an effect, they lied about it. Because the stigma was so incredible. So believe, by Share. They over did the auto tune. And there are articles upon articles where the producers

of that song just invented other lies to claim that they had done something that was not use autotune on Share because they were so worried about the stigma of having used the software. And now it's like everyone just uses it all the time. And I think we're in that moment with AI just like across the board creatively where no one wants to cop to it. But it's just going to be in the background of all these tools that we're using all the time. And the question is, can you make something that sounds different or great? And I think the worry is that you're going to type in pop song and get that and that will be all of music.

Yep. Okay. I feel slightly insulted. Just to be clear, we didn't prompt that. We actually worked to make something over about two hours that was songish. Songish. Yeah. Yeah.

You know. Using only AI tools. And yeah, it was very challenging to make anything. it just led us in strange directions. Look, I think that's as big of a hit as Believe. Oh, no, no. Let's get out there. Oh, come on. All right, y'all have a good day.

I'm out. So, Ian, if somebody walks into your studio and is like, I have the beginnings of something, I made it entirely with AI. Are you going to, like, roll your eyes and kick them out of the studio? Oh, not at all. I take demos from artists all the time. So a lot of times we'll do stuff from scratch, but generally I'm working with stuff who, like, they've already had the guitar part

and the vocal and that kind of thing. And yeah, there's no stigma behind any of it. You know, there's a bigger stigma, I think, right now, with using full samples than there is using any kind of AI tool, where if all of my drums are from one, like, drum sample loop, people are like, oh, you could have been more creative than just go on Splice and drag that in. That's more of a stigma than, like, using an AI tool at this point. That's really interesting. So, and what about the, like, I was just thinking listening to that vocal that has got to feel weird for you, having uploaded your voice to a piece of. software that spit back your singing in somebody else's voice. And I feel like I wonder about that creatively with all of this stuff, right? Obviously, there's big copyright questions about who owns a

voice and can you make a song with fake Drake that counts. But I wonder, like, it's got to feel weird to hear that out in the world, whether you're Drake or you, like, listening to someone else sing your song that was just your voice 30 seconds ago. What does that feel like? I mean, for me, I like my voice, so I'd rather just stick with my voice. But coming from, I work with so many vocal producers or producers in general, top line writers that do not like their own voice at all. And they have to manipulate it and tune it a lot to get to the point they would like to get it to. So if you're not a fan of your vocal tone, but you can get your point across in a recording, then you can change out your voice with something like this. And a lot of the work I do is I'll create a full beat,

I'll put on the top line myself, the lyrics and melody and everything, and then I'll send it to an artist. It's a pitch. And if they like the song, they'll remove my top line and put theirs on and keep everything else. So essentially, you know, you don't have to keep the AI generated voice, but it gives a reference for a pitch situation. Interesting. This is also just like an age-old issue. People feel uncomfortable with their voice. And there have been hacks that we've used for decades, whether that's reverb and compression or oftentimes, in fact, when you went in the booth, the very common thing to do when you're trying to improvise melodies and top line over a beat is you just put on auto tune almost to max and it lets you be looser and freer and so i think there's actually

something very yeah there's something liberating about being like i don't feel good about my voice but i know i can make a song and i can make it sound better if i just you know work and massage it with these tools neel i how do you feel about your singing voice yeah two three four uh you know when i was in the punk band in college i sang with a british accent that's a i that's a i right there that's just some straight al jurgensen did it in ministry and i'm going to do it too and that's that's That was the road I was in. You know, you said there's copyright issues. There aren't, though.

That's actually the thing, right? That your voice is not protected in that way. There's like a handful of laws at the state level that might protect it, but none of it's ever been tested. And the real question is not, can I sing into a computer and have it sound like whoever? So I can send it to someone else and they can. It's, can I distribute that and tell everyone it was Drake. Right. Or tell everyone that it sounds alike to Drake.

Yeah, I mean, I'm pretty sure we can't. release any part of what we made more because of a licensing issues rather than copyright issues because some of these tools we've used probably has some kind of license that needs purchased to be able to then, you know, use some of their sounds. And we did not do all of that perfectly. I promised David that I would bring up the thong song on today's episode of the Verge cast. What's fascinating about the music industry right now, and Charlie and I have talked about this lots of times, is that the norms inside the music industry about sample usage, about interpolation about all this stuff have almost nothing to do anymore with copyright law.

So I know this fact about the thong song now, and I will come back to it over and over again. Cisco does not own the publishing to the thong song because he said live in Lovita Loka in that song one time. Oh wow. No way. In the chorus. And Ricky Martin's songwriter came and took it and makes all the money every time that song is played. Wow.

That's just ridiculous to me. T. Payne, I think, just this week, revealed that he owns almost none of the publisher. to buy you a drink because there's all these little quotes in it and all those people came for him and they took his publishing away right wow and that's like they didn't go to court the norms that's just loyering up and yeah the norms in the music industry are just like we're gonna we've built this other system and it if you just described that to you it's like if you just whisper the phrase live in lavita luka like rickie martin appears and like takes your money away is that's not right it doesn't it doesn't sit right with me anyway and i think this ai stuff is going to lead to like a supercharged version of that where like the computer is automatically giving Ricky Martin money because it's been trained on his song like that seems like it will fall apart well it seems it seems to me like it goes one of two ways where either those norms become really really really aggressive and careful and everybody becomes very litigious about everything like you're talking about or it goes the other way and we end up in this place where it sort of nobody owns anything it's just

like pure anarchy and chaos because you don't know But we're going to get to this point now. I don't know if you've met the music industry. I mean, everybody's going to get religious at the beginning. I was described as a cruel and shallow money trench. But we're going through this with Chat GPT right now, right? Where it's like they're like, oh, well, this sounds like the book that I read or that I wrote. And Open AI is just like, who's to say?

And it's just like we're putting so many black boxes between I made something and something else exists that is kind of similar to the thing that I made. that I wonder if it's just going to get harder and harder to understand provenance and where all this stuff came from and what counts as a sample and who's responsible and where. And it's just like, I suspect the, when you say, how do you feel about this? This is the thing that gives me the most uncomfortable feeling because they're not going to not try. Right. And they're going to, they're already trying really hard. Like YouTube has a seal with Universal, which seems very terrifying to me. And they've already built this bizarre system.

where Charlie came on and we talked about Olivia Rodriguez, handing out publishing credits on an episode of Decoder, where Olivia Rodriguez is like, fine pair more can have a credit, even though probably shouldn't, right? The Taylor Swift ones are a little bit more. Yeah. And it's like the industry has decided they don't,

inside the family they're not going to fight. Right. And the famous people will just like collect some money from new artists. And you just keep spinning that up with AI and you're like, oh, eventually the rich are just going to get richer here. in ways that I think disincentivizes anyone from trying to make anything new. Like, you might as well, in that case, then just try to flip samples as creatively as possible and hand the money back out, right?

And that's the feeling here that it's like the music industry's response to AI is to become even more corporate in a way that just makes me like, like I'm an old punk rocker. Like, that seems like not what you should do. You know, like you should, like the kids should. revolt against all of that. And like, Olivia Rodriguez should be like, I'm definitely not paying you. But in the same way the industry, as I've noticed, is getting a lot politer in some ways. And I realize when we think about that, it might not look that way. But, you know, when I have writing sessions with people now, there'll be four people in a room. Let's say three of us do 90 percent,

not 99 percent of the work. And the other person lays one base note down. That's four equal splits on publishing. And I kind of see it the same way with, you know, you're anticipating the lawsuit. and there's going to have to be a payout regardless and all of this stuff just to make it go away. Like there's definitely that side of it. That's why you add the credits initially. But it's more so just like a, I don't know, a peace offering of kindness to be like here, take this.

And it is cool seeing people like, you know, higher up artists that are handing out to like lower level artists, these publishing credits and that kind of thing too. So I don't know. The way I think about it is getting politer. But yeah, when we're thinking about AI in the industry, it's corporation. Yeah, and then it's automated.

That's the thing that gets me. It's like, okay, some artists are handing out some credits as they go through a writing session. Okay, like, I don't know how you feel about Ricky Martin, but like, yes or no, that happened, right? On the flip side, it's like, okay, I made a song using AI, uploaded it to YouTube. YouTube is like, Drake gets a piece of this. Yeah, I mean, I don't make Switched on Pop video versions for YouTube because the content ID system is such a pain that I don't want to deal with it because it will demonetize everything. you have to fight everything. Even for journalistic purposes, if you use any copyrighted material, which is absolutely fair use,

YouTube just as its own government, right? With its own set of ideas. It's not actually a copyright issue. It is a set of private contracts between copyright holders with YouTube that has a system that you have to go through that is its own court. And it's just such a pain that I don't deal with it and I'd rather just make a podcast. Right. You were mentioning before, like you didn't do this process of making all this stuff in the most sort of like

upstanding pay everybody who should weigh. But like you left a trail that every one of these services knows about, right? And to your point, as they get better at doing things like having deals with record labels and stuff, like, it may be true that that money just like slowly starts to disappear every time you open a new app. And right now this stuff is happening like in discords. And like one of the most popular things is like a, it's in a Google Drive and you just upload your file to a Google Drive and then download it from a different Google Drive.

And it's like, that is fine and going to be hard for anyone to figure out how to make money off of. But eventually, if this stuff gets better as quickly as people think it's going to, it's going to get corporatized too. They're going to make these same deals. And then all of a sudden, as you make music, you're just going to leave these like red crumbs. A number of these services claim to have watermarks within them so that they could identify automatically with systems if you haven't properly licensed. Do you believe them? You know how like the BBC used to drive around London with a truck that could detect. tech TV antennas if you didn't pay the fee.

I'm just saying, man. I think the question is, how good do AIs get at tracking those things down for them because it's a pain to enforce? But I think enforcement of copyright will get significantly easier. I mean, they're people making very effective DRM tools that can basically track every misuse of copyright on Instagram. People just, because Instagram is basically a one giant copyright cluster, right? People are taking people's intellectual property all the time and reposting it without permission.

and no one's really enforcing it because it's difficult, but there are companies that can help you do so if you want to. But it's also mind-numbing because, so I put out a song that was 100-per-per-per-per-per-I-wrote it. I wrote it. I did everything on it. I had 100% of the master 100% of the publishing. I put a TikTok video out using that audio, and the audio got removed for copyright. No. What?

I own it. Yeah. Yeah. But this is, so like all these automated systems have some error rate. Yeah, right? And that error will just take. Did you have any, did they tell you? What did you do wrong? You literally violated your own copy.

Exactly. The estate of Marvin Gaye shun. This is why actually we need AI music. We can't make music anymore because it's so litigious. This is what I'm saying. You build these systems with this weird understanding of who gets paid and why inside of it. And the answer is because you exist and you might sue us. And then the error rate in the system makes it so new artists can't upload their own music.

And it's like, well, that's just definitely the wrong outcome. You make, I'll give you the blurred lines example. Like you open an AI drum program and you're like, give me a Marvin Gay beat. Yeah. And the state of Marvin Gay is like, that sounds too much like any song. Can you imagine that where you could prove that AI made it and then you can like counter sue? I just think like that's the thing, like that's the feeling I get with all this stuff is once you just add this much software to the creative process and the pressure on the software is to do automated enforcement. or automated credits or to not do things to protect you.

It's like, oh, we've just pushed creativity into one tiny little box. I honestly, throughout this whole conversation, I'm still wondering, like, what is even the point of using any of these tools? Because in music at this point, there's just no scarcity. If you want a sound, it exists. You can get it for free or cheap. If you want professional-level tools, you can use them with a decent computer. There are free tools available to make very high-quality music. Like if you asked us to go do this process again using only $100 guitar and plugins that were free off the internet, we would make a much better soul.

No, just BanLab.com entirely free and anyone can use it. I guarantee you just pick loops and you drop them in and it manipulates them and makes them all fit together. I've taught BanLab to eight-year-old children and they had it like that. Wow. And it's so, so easy to use. And it's free. And everything is yours. It's entirely royalty-free.

So why? But we don't need these tools to make music. for us unless there's definitely like the amateur who's trying to learn or wants to express something I think is certainly interesting. But as, you know, working professional, it's kind of like if you want the high quality sound, you're not getting it from these things. So why even like deal with the future headache that is discussing? You put yourself in that world.

Yeah. This is just at the bottom of the curve, though. I mean, that's kind of, right, this is the 808 example. Like it didn't work for that workflow and then the stuff got discarded and people who were not part of the gate cap system. We're like, we're going to use it for our own thing, and that became a dominant sound. Like, that's the pro case here. There's also this idea that culture radically shifts and that all we want is completely individualized art, which doesn't seem very human.

But perhaps that happens. And then, like, yeah, we all make our own songs that are in our own perfect little world. The thing is right now, like, already there is so much high quality library music, you know, like better pharma sounding music that you could go listen to, which is like your own. own special personalized playlist. And actually that's not what people usually want. They want to be in conversation with culture what other people are doing. And so that world doesn't really make sense to me. And it goes back to something you were saying at the very beginning, which is like if

you, if what you want is an easier way to make like a music bed for your YouTube video, tools abound and they're getting better very quickly, if you want something that is like good and high quality and will make you feel things and care about things and like be on the radio, we're kind of nowhere. I still think these tools kind of, they're pretty janky most of the time. Like, we just did an episode, you know, Jimmy Buffett recently died. And so we did a little tribute episode to him on Switched on Pop. And I had to make a promo for that episode.

And I was like, can I just find Jimmy Buffett-esque music? And I couldn't find anything good on Splice. And I went to some AI tools and was like, make me a Jimmy Buffett song. And I was just like, screw it. This is so annoying. I brought out my acoustic guitar and I just strum some chords and it sounded way better than anything else in the genre. Well, when Jimmy Buffett. dot AI launches. We're having you all back

and running for seeing you. But until then, thank you all for being here. This was unbelievably fun. I really appreciate it. Charlie, will you play us out? Yeah, run it back. All right, that's it for the show. Thanks to Charlie, Ian, and Nelai for being here. And thank you, as always, for listening. AI David, do you want to hit the credits again?

If you have thoughts, questions, feelings, or other songs you'd like to hear AI David sing, you can always email us at Vergecast at theverge.com Or keep calling the hotline. 866 Verge 11. We love hearing from you, and the meta-vergecast episode is coming soon, so if you have questions about The Verge or The Vergecast, send them in quickly. This show is produced by Andrew Marino and Liam James.

Brooke Minters is our editorial director of audio. The Vergecast is a Verge production and part of the Vox Media Podcast Network. We'll be back on Wednesday with some more Apple Talk, and with plenty of new news on Friday. See you then. Rock and roll.