The Illiac Suite - Music and AI

This is an automated AI transcript. Please forgive the mistakes!

This is the Iliac Suite, a podcast on AI driven music. Join me as we dive into
the ever evolving world of AI generated music, where algorithms become the composers
and machines become the virtuosos. Yes, this music and text was written by a
computer and I am not real, but... I am and my name is Dennis Kastrup.
Hello humans, welcome to a new episode of the Iliac Suite.
I was on a mission the last weeks and looking for the most interesting songs
created with the help of an AI at the moment or in the past weeks.
Between you and me, I got a little bit lost in this wide open space. There is so
much happening out there. Hard to put it into one podcast, but I try it.
However, I hope I can offer you a little glimpse into that world by playing some
songs and also listening to some of the creators and their thoughts on the music.
I talked to two musicians about their approach when they create music with the help
of AI, but this episode does not work without a disclaimer.
Here we go. What you are about to hear are songs I found online. My research
showed me that none of these songs are part of any collecting society worldwide. It
is also difficult to reach out to the creators behind the music. Sometimes they do
not answer, sometimes they leave no traces, and sometimes I do not even know which
part of the music was written or played by them or an AI. So, we are probably
currently in the greyest zone ever in the music industry. Nothing has really been
sorted out about music in data sets, AI voices or songs written by AI. Welcome to
the Wild Wild West. If you believe your copyrights were infringed by me playing your
song, please reach out to me before talking to your lawyers. So far, nothing is
legally ruled anyway. We are in this together. I do not have an idea how we will
solve all these problems. I'm just trying to find some positive aspects of it.
Thanks. So let's do this. What you hear in the background are the first beats from
a song called "Banshee in the Rain" by an artist called Werthein. The voice is sung
by Grimes or better the AI voice of Grimes. She is probably the most known
supporter of AI voices right now.
I heard crying, I
heard crying,
mutin' (upbeat music)
The user we're tying with Banshee in the rain. I don't know if I pronounced it
right W E R T A Y I N. I found the song under the hashtag Grimes #grimesai
songs. It was made with Uberduck, the AI with which you can make famous voices,
sing, rap or speak. I guess it is one of the most used AI in the music right now
besides voicify when you want it the easy way. The more complex way use the open
source software called SoVitzSVC, singing voice conversation.
You need to code a bit, so there is some effort involved, but the results are
stunning. You will find all the information about that model on Github. The song we
heard was part of the competition to create music with the voice of Grimes. This
one did not win, but I liked it, because to be honest, not a lot of the songs
were good. Unfortunately, I did not find out who the user where Tyne really is and
how he or she made the song, but I found an old quote from Grimes when I
interviewed her for her third album in 2012. The closest statement from her on what
she is doing today, wanting to push the frontiers in music with her approach to let
others use her voice and getting 50 % of the royalties of the song, is this
following quote. When When I look back on what is humanity, I'm like, "Oh, it's
like Da Vinci and Shakespeare and like, I don't know, Lydia Lynch and like, just
anyone who partook in an artistic, cultural dialogue and that's like what I want to
do. Obviously that's a really ambitious goal or something and I'm not saying that
I'm anywhere near as qualified or as good as those people are, but I mean just
even trying to do that is like, "What else are you going to do with your life?
I'm not of her bit I would ever have a job again like I'd rather starve than like
work in a restaurant. I guess she reached that goal with being together or sometimes
not being together with Elon Musk. Here is the song that won the Grimes competition
born to destroy To destroy,
tell me What have I done? Was I just a toy?
Obliterated all within my nature And now I regret The moments that I realize I I
realized I killed who I loved Then you since you exist broke me I played gore and
I lost
Why did you do it? Who are you?
Those Hopes and dreams and machines Wishing it wasn't true Old literated all was the
nature of Now I regret the moments that I realized I realized I killed Who I
loved, that you since you exist, broke me I played God,
and I lost Now it's gone, it's gone, I'm alone Wish I knew what I,
you created me Knowing I, didn't ask to be bored born to destroy with the I voice
from Grimes from Max Weisel and Juliana Juli Ronderos.
I talked to Max, who is an American software engineer and digital artist. He
explained me how they created the song. I've always worked with music and computers.
I had previously done some, like when I first got my start, which is maybe like
12, 13, 14 years ago now, I was doing a lot of like small iOS instruments and
things like that. And I actually ended up getting hit up by Bjork. And so I got
to work on the Bjork Biophilia album. And so I made a bunch of apps for that. And
some of it even got used for composing songs on the record. And then Since then,
I've delved a lot more into the tech side of things, but music's always been
something I've been very passionate about. When a lot of the AI music stuff started
happening, I had actually been slowly chipping away at a different AI music project
for a while.
I saw the Ghostwriter song that was making waves on TikTok and looked into that and
saw how I was that a Soviet's SVC voice changing AI stuff, which basically is like
deluxe tamper transfer for voice. And it was really cool because it was essentially
you could, there's something to be said about the familiarity of someone's voice or
like a vocalist, you know, and so it was kind of cool to be able to write a song
and use a vocalist you're familiar with. And that ghostwriter song was such a great
example. Like, As far as I know, the whole song was written by a person and he
even recorded his own vocals and then used that neural net to change his vocal to
sound like Drake and The Weeknd.
And so like, you know, you have to be able to write a good song to do it. And
he had written this amazing song and in a way like no one would probably have
given him the time of day to listen to it if he hadn't used those voices. And
it's And it's funny, it's like, you can write a great song and in a way it's like
80 % still marketing and just getting it in front of people. Funnily enough, so pure
coincidence when I was, when that came out, I happened to be in LA and I saw
Bjork there and I asked her about this and it was interesting 'cause her kind of
take was like the tech is really interesting but your voice is sort of like a
passport, you know, it's your identity in the world of music. And so allowing anyone
to just use that is a little bit, a little bit scary. And so, uh, and I totally
agree with that. I mean, it's, it's interesting to think about how things are going
to change. I mean, in the pop world where you've got pop stars and a room full of
producers who compose and produce and do everything and the pop stars come in and
lay down the vocal, I've certainly, it's not everybody, everybody, but there are pop
artists like that. This would allow them to potentially at some point not even have
to go into the studio, which I think is kind of an interesting dynamic. But yeah,
so I was curious to play with it myself and I saw that Grimes had said that she
was gonna do a 50 /50 split on the master if you made a song that became popular.
And so I was already kind of looking at it and then this UberDuck competition came
out And I actually only had like a weekend to do it, like a day and a half.
And so I sat down, tried to make something quick. My friend, Huli, who's in a band
called Salt Cathedral,
she, I've been talking to her about this stuff for years. And so that weekend
rolled around, I was like, Huli, you need to help me. I can produce the track, but
I'm not good with lyrics. And so we just sat down and in a day basically wrote
that song and Huli wrote all the lyrics and I did the melody and the production
and we kind of just slowly noodled on it until it was perfect. And yeah,
and we went for a piano ballad because I was like, cool, I got to do a piano
baseline. That's about it. I really wanted to do like a really cool grimes, like
distorted guitars type song. But But yeah, I just didn't have the time to do it.
Yeah, so anyway, so we did that in a weekend and put it out. I mean, for what
it's worth, I do not think objectively we were the best song. I think part of it
is it's somewhat of a popularity contest and posting it to Instagram is a good way
to get a lot of your friends and fans to go vote for stuff. But that's kind of
how I got into it. - I guess one thing we should consider these days is that in
my opinion, it is still not clear to many people how AI music is made. They hear
AI music and think pushing some buttons makes a good song. That is definitely not
like this as Max just told you there's a lot of work on top of it also for the
voice. So I tried it originally and my vocal register is probably too low but so
Hooli sang all the vocals and then we them to Uberduck and used,
I actually used the RVC model. I thought it sounded better and it's kind of wild
because at first Huli I think kind of just sang in her style and the output was
pretty amazing. Like it even adds like the little like there's little details in
Grimes voice that really make it sound like Grimes that it would add. It would even
add Lisps here and there occasionally And then after playing with it,
I think who we started to try and almost do like a Grimes impression. You don't
have to do it too much, but maybe just in the way you sing and in the way you
change pitch. But with that, it's like you get your little wave file,
you upload it to UberDuck, they run it through that SVC or RVC model, and outcomes
almost the exact same clip. Funnily enough, same clip. Funnily enough, all of the
clips came out a little flat, and so I had to auto -tune them, and I would sit in
Melodyne and tweak them. Same thing you'd do to any vocal. You're going to compress
it. You're going to tweak it a little bit so it sounds to get the best pieces to
put together the best vocal performance that you can. But for the most part,
you press a button and it turns it from one person's voice into another. So yes,
there are AIs where you define the patterns by just clicking and music comes out
also for the voice, but normally that music is not that good, right?
Means for me, there will be no good AI music out there if the musicians producing
it are not already good musicians who know how to make good music. That's my
opinion. Well, the most known example for this so far is Cold Touch. The producer
Kaito provided the beats and the Scottish vocalist Nina Naspid then fed through
Grimes' new AI platform, Elf Tech Her Voice. Nina is a professional singer whose
voice is not too far away from Grimes.
You could be the one that made me cry Lay me up inside,
inside I'm sleeping in your worn out sweater I'm just trying to literate the signs
Won't you tell me why? Oh why? Blue blood cold touch You could let me all night I
shall switch up talking to the war way I'm gone, I'm gone You're out of your mind
I can't feel the breeze, baby I am on a sea,
I don't love me
I can't feel the breeze
I just want the real thing to feel I want you to hurt me too only close Take
what you need, get your done And then you just wanna make me scream Don't look,
don't touch, you're gonna be all right I just will jump
Cold touch by Kaito featuring the voice of Grimes featuring Nina Nesbitt from
Scotland. I have for you another AI voice, a cover of a Turkish singer called Tuce
Gandomir. Sorry for my pronunciation. The song is called Yanlis. The original has 144
million views, I guess a mega hit in Turkey. Let's play a game. Which famous
American singer is that supposed to be?
Sadece baştım, baştan acı günleri Sor ama belki tek beni iyi ekeceğiz Sorma sarmışın,
aşkı saklı dünleri Beçledim günleri Elbette Gece Her şey uyumda Kime gözüküyor oradan
Bırakacağım temana Geri deyip anana Dayana dayana Bittim çok sen Dedim ki vardı
Yerime yamala yıpsın artık San derman derken Fark Benim yoruldum bu aşkı Benim gözüm
ama yaşlı Anlatamadım kaçtım Başlam yanmıştı Yoruldum bu aşkı Benim gözüm ama yaşlı
Anlatamadım kaçtı Başlam yanmıştı
Saati cevaştığım, baştan acı dünle Zor ama belki,
beyni eceç Soğurma sarmışım Aşkı saklı dünle Beçledim günle
Bekleyecek Ayşe yolunda Gibi göz öpüyor oradan Boyaktam sen bana Geri de kalana
dayana Ben habittim çoktan Didim ki vardı Yere bu yaman ayıpları Sanderman belkere
Fark ettim Yol uzun aşkı Benim gözüm ama yaştı Anlatamadım taştım My head is on
fire, I'm on my way
My eyes are on fire, I can't explain it My head is on fire
The musician who is singing here in Turkish is supposed to be Rihanna. I get it,
but I still think the voice is far away from the real Rihanna and you can hear
the glitches also sometimes in there. I reached out to the uploader and profile
called "Miksbuni"
because I wanted to know how he did it, but unfortunately I did get no response.
He must have had a voice singing it? Or did he have a clean voice of the original
song? I will unfortunately stay in the dark. There is also another Ariana Grande
song in Turkish by the same uploader Miks Buny. If you are interested, check it
out. By the way, thanks AI for showing me the beauty of Turkish music.
Talking about beauty, the discussion about AI and music has been speeding up lately
because of a user named Ghostwriter and his album Sincerely Aubrey, a new full Drake
album with the AI voice of Drake and others like The Weeknd, Kanye West or Kendrick
Lamar. My thoughts on this. The album is good. The music is good. Are the voices
all not real? Could it be? Could it be AI? Yes, but it could also be just real?
It is so confusing these days. Or is it a promotional stunt? What speaks against
this? The reaction of universal music, which took it down, seems like they were not
amused, although they argued it was because of a sample in it. The beats are great,
it is good produced, so who is ghostwriter? I don't know, we prepared. This is just
the beginning of confusion, because theoretically in the future stars could also sing
and claim the song was made by an AI, or the other way round, I mean, these AIs
will get so good. But in the end, is it really important to know that a song was
sung by an AI or not? If it touches you, it is your song,
no matter who sang it. Okay, concerning copyrights and royalties,
it is important who sang it. But who gets them anyway? The owner of the voice, the
producer of the song, the programmer of the AI, the musicians in the data sets?
More about these problems in one of the following episodes here on the Iliac Suite.
So again, is it important to know who did write a song? I say no,
not for the music pleasure. Take your song, do whatever you want to do with it. If
you like it, if you love it, just dance and shout to it.
And to prove that, I have my friend Sarah Mackenzie from the famous Mutec Montréal
festival telling us now a story about what happened to her in New York a couple of
weeks ago. Sarah. This past April, I experienced a very uncanny and surreal nightlife
experience featuring artificial intelligence as a main character. I was at a party in
Brooklyn and the night was coming to a close and all of a sudden the DJ puts on
this track and the crowd absolutely explodes. And what was that track?
Well it was the AI created Drake song And I absolutely just couldn't believe the
reaction of people who were there. People were dancing, shouting the lyrics, having
emotional reactions. It was absolute euphoria, and it was just as if it was an
actual real Drake song. Which kind of begs the question, do people even care if
it's really Drake or not? To be continued. And of course, here it is, hard on my
sleeve.
(upbeat music)
Okay, maybe I don't play that song too well, because of some reasons you heard
earlier on the show. One thing that is obvious in these days of music is that AI
creates a lot of rap music. To understand how that works, I reached out to Matt
Chu, who you can find under the profile "Mellow Flows", that's his artist name.
I talked to him and that's what he says about himself. About 15 years ago,
over 15 years ago, I started writing raps because I joined a rap message boards
online and this was strictly for battle rap. So they would organize tournaments and
online battles where I would be set up against another emcee on the boards.
We have a couple of days to write and then we record something on our own ends
and upload it to various hosts. And we would listen to it.
Others would listen to it. People would vote. And That's really how I got my foot
in the door as far as hip -hop goes. It all started for battle rap for me so I
always held like punchlines and complexity and that type of stuff as in high
importance over like pop and Vibes and that kind of stuff and and yeah My brother
as well He joined me in that enterprise and he started making beats and we just
started making our own songs together, it kind of evolved into a lot of different
genres. Mainly it was like gangster rap and trap music or like old school type of
thoughtful type of messages. Met Raps and his voice is transformed into the voice
for example of Jay -Z, the notorious B .I .G or Tupac. I really want you to feel me
on this
I find myself thinking about my father drinking missing up at the house My mama
kicking me out, I look at them and me now What do I see having children's about?
I'm one of many trying to figure this out and I've been known for dintin' around
You might've heard me diggin' it out with your bitch on the couch I'm tryin' to be
responsible for the kids who I failed to middle ground Bitch, take the stick in
your mouth, I'm just a man, I admit it The plans on bitchin' is evolving faster
than the man With the spirit, guess I am quite ambitious, I'll be back in a
minute, girl I'm tryin' to rule the fuckin' world, don't you know I'm just a man
I'm
just
♪ It's obvious I came a long way, gotta hand it to me ♪ ♪ Murderous, we'd ever
be clear, like a pack of Newsy ♪ ♪ This is how sun sets in, burning Cali do me
♪ ♪ After all that's happened to me, now you can't refute me ♪ ♪ One of the
braids, simulate one of the braids ♪ ♪ Did you get one of the braids, still a
witness in greatness ♪ ♪ I'm just trying to be myself, if that makes sense ♪ ♪
Baby, have patience, I'm just a man ♪ - Melo flows with just the man from Tupac by
Tupac's voice with Tupac's AI voice. I think we have to find out what we'll say in
the future. I wanted to know more about Milo. So how did he write this song?
Yeah, I was just curious about how it was done. So I started doing a little bit
of research and I found out that it's extremely accessible. Just with one link,
you could get access to a Google collaboration session. And right there,
they have this really messed up interface. It's not very nice to look at, but it's
pretty straightforward. You just connect to it. You upload your audio files,
which is your vocal files, to your Google Drive. And you share access to your drive
onto the Google Collaborate session. And then from there, the tool can actually grab
your files and you choose through like this ugly dropdown,
one of their many, many voice models that they have currently available.
And it takes quite a while and it'll time you out if you're too slow also,
if you have a ton of But you you choose one of your files you choose the output
model and There's a whole bunch of little settings that you could tweak and what it
spits out is The the the model trying its best to convert your voice or whatever
you give it Into the artist voice so you could just be blabbering saying nonsense
But it would come out in the target voice. Here is another example,
the notorious B .A .G and Eminem together, I wonder.
Ain't enough, I'll be the first one to knuckle up Kiss my ass, bitch, pucker up,
what the fuck, dumb ass slut Don't touch the drunk, shoveed up my butt I was
saving that for later, but later is now Mayors will now let you play this shit out
Don't talk about me, gettin' reckless, crazy and wild That's what you wanted, isn't
it? Well, then say that shit loud I've been waiting to cut loose again The future
brings whatever the fuck, I'm not this boost I'm in, if you My enemy then you
refuse to win, I keep it one -handed just like the proof I drink It's too easy to
rap it, I keep telling them we fellas in, still getting paid Benjamin's Fuck haters,
drink the hell while we sending them, sometimes I wonder if this gang life ever
ends Mellow Flows rapping with the voices of Eminem and the notorious B .A .G.
Talking one more time about the voices Because that is what got me interested in
mellow flows, Matt says on his profile that with rapping he feels that the voices
of others help him encourage to say things he wouldn't normally say.
That is very interesting and I wanted to know more about this. Okay, the beats, I
just downloaded them from producers that uploaded it to beatstars .com.
I just go there. I search for free beats. So those are free to download free for
any promotional use. So I just download them, whatever vibe I get from them. I feel
like, oh, this artist would sound great on this beat. So I then write the lyrics
from my own perspective. These are all my own perspectives when I write,
but with the mindsets that this artist will be saying it for me kind of.
So I try to pick a flow, right for a particular flow that meshes well with my
target voice. So yeah, everything I write, you know,
I don't pretend to be them. I write it as myself. It's just being said through
their voice eventually. With a voice like for example, Tupac, I just feel more
comfortable saying certain things. Of course, I don't feel like that gives me a free
card to say the n -word and that type of stuff. But just to be more gutter,
to be more vulgar, I could use different voices and I feel like that allows me to
get away with it more. Because if it's just my voice, I don't know, it just
doesn't sound Like, I should be saying those things, but if it's from their voice,
it's almost expected and very much more acceptable. It's definitely more free,
I feel liberated. And in the song that I made with the Jay -Z voice,
that's exactly what it's about. It's called "The Ruins," as in, like, this is the
ruins of music, of creating music. Like, everything is ruined,
basically. Like I have the freedom to make something that I want to hear. And if I
want to hear Jay -Z saying this stuff, well, I have the resources to do it. So I
want to make that kind of music. And that whole song is basically talking about how
myself, Mello, can write this, write these bars from my perspective,
but have another voice say it and it's completely limitless.
I'm not tied down to any particular voice or any kind of limitations in that sense
at all. And I can make artists collaborate that would never normally collaborate.
Of course, we could bring artists back from the dead this way. And it's just super
Deliberating like I feel like the the it's endless
This is spiritual lifting, this my R &R Non -stop throw me beats like Mardi Gras I'm
so limitless, not definitive, not finite This the high life, and this is my right
Shit that I write, shit that I lay down Spit in my mic, I might be a freaking
nature now Don't know if I can make it out I make a sale then I take it out Can
I still say something that they gotta think about Fuck that I mean, is it really
my fault I got the skill and the resources to
Mellow flows with the ruins in the style of Jay -Z, I think it's actually pretty
good the lyrics the beats and the rapping the original rapping or the rapping of
Jay -Z I don't know make your decision and here I would like to end with some
thoughts for you pop music is full of stars who never wrote one lyric or note by
themselves Elvis was the first example of that he just took ideas and people of
songs for him. What you see in the modern pop music world these days is full of
autotune, ghostwriter, I mean not the Drake one, so real ghostwriters who write the
lyrics for others, producers who write the beats or even buy them for others.
I worked for a big major label a bit and I know that this world was always fake
and will be. Do not believe the stories they tell you. They just want to sell and
make money. All this business model is now in the hands maybe of one person at
home who can let their biography be written by JetGPT. Music also and lyrics can be
made by an AI. And for visuals, you have mid -journey, dolly and so on.
But You can do this as much as you want. If you are not a real musician by
heart, a real artist, you will never succeed with AI music because in the end, it
is you who decides what to do with it. AI will stay an amazing tool for creative
people, but the question remains, who will get paid for that? I have no idea so
far where this will go. I will keep you updated. Maybe in some years, I know here
on the iliac suite what to tell you I just hope that everyone who deserves it to
get paid gets a share of the cake Say it in other words all of you Get lucky
whatever you do sung by Frank Sinatra made by the AI hit factory using voice
cloning technology and the new tone orchestral song "Get Lucky" because that's also
AI music, a pool of foolish, stupid and sometimes a bit funny ideas.
Sinatra style Like the legend of the phoenix
I'll land us with the beginning
Get the planet spinning
The force of the beginning
We've come Too far to give up who we are Too late to raise a bird Tell her God,
tell her star
A simple night to this sun A simple night to get some She's a simple night for
girlfriend A simple night to get like me A simple night to this sun A simple night
to this sun
Rock ball night to get, Rock ball night to get lucky
That was episode 2 of the Eliak Suite.
I hope you enjoyed it.
Thanks for listening, humans. If you wanna live, I'm ready Take care and behave
We've come too far to give up who we are Let reason burn,
get our God to the start He's up all night to the sun,
he's up all night to get some He's up all night to go far, he's up all night to
get lucky He's up all night to the sun, he's up all night to get some He's up
all night to get far, he's up all night to get lucky He's up all night to get
lucky, he's up all night to get lucky We're up all night to get lucky.
We're up all night to get lucky. We're up all night to get lucky. We're up all
night to get lucky. We're up all night to get lucky. We're up all night to get
lucky. We're up all night to get lucky.