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'm not real, but... I am and my name is Dennis Kastrup.
Kiss me up before you go
Summertime's the last I just want you to know
I got my red dress on tonight Dancing in the dark in the pale moonlight Some more
hair, real, babe, feel it when it's down Hot heels off,
I'm feelin' alive
Oh, my God, I feel it in the antelope, I'm wise I'm
*outro music
*
♪ I just want you to know ♪ ♪
I'm I'm sitting in the electric tonight Cursing down the clothes,
going 'round on your night Got my back, baby, by my heaven residing I know if I
go, I'll die happy tonight
Oh, my God, I'm sitting in the air Tear up from wires above I'm sizzlin' Like this
man, I'm there, oh, I'm fine, I'm feeling it
Before, kiss me all before you go And
the times that you live I just wanted you to know That baby,
you're the best I got that summertime, summertime sadness Summer,
summertime, summertime sadness Summer, summertime, summertime sadness Oh,
oh, oh Did you ever miss me forever? Like the stars with the sun in the morning
sky Oh, oh, oh
I got that summer time, summer time sadness Summer time,
summer time sadness Got that summertime, summertime sadness
Kiss me already before you go Summertime sadness
I just wanted you to know That baby you're the
♪ I got that summer time, summer time, sadness ♪ ♪ Summer time,
summer time, summer time ♪
"Summer Time sadness sung by Kanye West, so it is proven Kanye is not only a good
rapper, but also a good singer. Of course, this is not the real Kanye West, but
the AI voice version of him or better, the AI that changed the existing Lana Del
Rey voice into the famous rapper's voice. The profile who put this online on YouTube
is called AI Cover Songs. He or she was pretty active three months ago,
but something must have happened. The linked website aicoversongs .com doesn't exist
anymore. And there are also no more new videos since song weeks. No information
about the uploader. Did the uploader get bored? Did something happen in his or her
life? Or one other thought was the fear of being sued by the people behind Kanye
too big. Who knows, as it is impossible to reach out to the uploader,
unfortunately, I cannot tell you what the reason is, but there is one thing I can
tell you. It seems like in the ocean of uploaded AI generated music,
YouTube does not do too much about it at the moment. Although a lot of these
videos claim in the title that they are made with the help of artificial
intelligence, they are all out there. So I do not get their policy so far. They
take down hard on my sleeve, although there are also many new versions of it on
YouTube now, but then it is still the Wild Wild West with other videos and AI.
It's a weird policy right now. Was Universal Music such a strong partner that they
could argue it was copyright's infringement, although that field is not regulated yet.
It's a bit grey. Something to find out in the next episodes of the Ilyak Suite.
Today I'm pretty sure that we will not have any problems with that. I talked with
the talented and amazing artist Portrait XO, who was besides Holly Herndon,
one of the first persons to experiment with the own voice generated by NII. To
introduce her, I have my synthetic friend, let's call him Jonas, by my side who
will tell you more about her. Portrait EXO is an independent researcher and artist
who creates musical and visual works with traditional and non -traditional methods. In
collaboration with Databots, they won Best Experiment Award at VUT India Awards 2021.
Eurovision AI song contest jury vote for most creative use of AI in 2020. She
researches computational creativity, human machine collaboration, and explores new
formats and applications for forward -thinking art and sound. Her debut research -based
AI audio -visual album, Wire, was released on December 9, 2022. Her obsession with
creative AI has given birth to new methods and approaches to songwriting, sound
design, installations, and storytelling. Co -creating lyrics and melodies with an AI
version of her voice, they performed together as a neural vocal duet. The biggest
contemplation with majority of Portrait Exo's work questioned what it means to be
human in a data -driven society. Thanks Jonas, so let's listen to her music.
Jung by Portrait Exo.
Jung by Portrait XO which by the way means Young in German and attachment in
Korean. What I find really fascinating about this song is how Portraitix O really
goes hand in hand with the audio files generated by the AI. It is a joyful play
that creates musical beauty. But before I put too many words in her mouth,
I'll let her speak about the process of working with AI. - I guess for the last
few years, I've been deeply researching the different ways that AI can be creatively
inspiring. I started first with music, collaborated with CJ from DataBots,
who I then met Zak Sakowski and them too, who are both musicians and data
scientists from DataBots. And they've been pioneering this whole area of AI audio.
And I guess CJ introduced me to what he was working on back in 2015 and at that
time it was even more abstract compared to where we are now and yeah he just kept
encouraging and poking every now and then like we should try some experiments and at
the time I was kind of going through a really really tough writers block I wrote
and recorded like over 200 songs in a year and I wasn't happy with like majority
of the stuff. And I was desperately trying to find new ways of approaching the way
I wanted to create as well as, yeah, I just wanted something to shake me up and
take me in a direction that felt new and different and interesting and inspiring
again.
In 2019, CJ and I both happened to be in Berlin and we both got into Factor
Berlin's owner plus D artist residency. And that was when we finally got to
collaborate. And I gave him a whole hour recording from those 200 songs that I
didn't release. Kind of like cherry picked the best audio, the best vocal takes.
And then I gave that over to CJ for training and then he trained it for two and
a half days. I had no idea where any of this was going to take me. And that's
exactly what I wanted, which is what ended up just being such. It's been maybe the
most imaginative, re -imaginative album that I've made. And for that reason,
it's been really exciting to share it because it's been mainly about like the
workflow and the method and the approach that's been really exciting to talk about.
And yeah, so I guess what came out of it was this co -creative process of finding
these little gems of tiny little bite -size audio where they had this melodic
information. Um, but what was interesting was it wasn't like fully generated phrases
and like really long, um, melodies. They were really short, like three seconds,
five seconds. And so I ended up curating a bunch of this content and then started
co -writing with it. And it became this like fun game of call and response. So it
would say something and then I would fill in the rest of it. Um, and what ended
up starting as a very scientific approach to co -creating with AI ended up being
quite a personal and emotive journey because it was just so like the,
I just remember having so many visceral goose bump moments, just hearing my voicing
back in really strange peculiar ways. Cause it was combining vocal techniques And
then singing words that were like very, I guess, profound to me because they were
things that I was kind of contemplating on, like love is misleading, you know,
hearing my voice laugh at me and then sing love is misleading is just like,
you know, these kind of moments when I just found myself like stopping everything
I'm doing and taking a step back and and staring back at like what just happened,
you know? - And how that song sounds, we will hear now, love is misleading, listen
closely, because all these AI audios are of course an interpretation from PortraitXO.
So theoretically, you can hear something
Cause love is my state Shall we go somewhere Lost
on the state Can't continue Just live here
I'm in love
There's nothing ♪ To tell if we're too old ♪
That's me
That's me Ha ha ha ha ha Ha ha ha ha Ha ha ha ha Strip out of that life The
things we keep on the stone with love right now We end up with some half love and
not worth the time we were losing High,
high, it's a lie Love is my state and you say ♪ Love in which you can't keep
touching ♪ ♪ That's why love is ♪ ♪ Sleeping ♪ ♪
Breathing
deeper ♪ ♪ You're here when a child in a dark room ♪ ♪ With deeper thoughts you
never knew ♪ ♪ On a way of home to memories ♪ ♪ You hold a place of love you
can't see ♪ ♪ What's real love ♪ ♪ Love is misleading,
you can't keep on touching ♪ ♪ Thoughts running around and around again ♪ ♪ That's
why love is misleading ♪
The discussions about data sets and the music in there are still going on and
probably will go on for a long long time It is not really clear on what data
Google's music alarm and noise to music actually was trained Same thing for other
data sets. So this is still tricky unless you train with your own data We're
currently going through this like mess of trying to understand policy and regulation
right like copyright related to AI generated stuff and so I was really intentional
even from the beginning to make sure that I wanted to use my my own pure datasets
of stuff that I created um and so when I get asked from musicians who are not
super tech oriented like, you know, they're, they're, they don't,
um, record their own music or like, you know, they're not really into technology in,
in that way. Um, I, I try to encourage them to at least educate themselves on what
ways this technology is impacting their role as a musician. And,
um, the data set determines the output, right? So, Um, as I've gone through the 200
plus songs of recordings and, and sifted through with the fine comb to find only
the, the bits that I love the most that I felt proud of to be used as training
data is the equivalent of, you know, if someone wants to become the best drummer or
the violin, the best violin player or whatever, it's like, yes, go and do that. I
mean, it's like, I don't think any musician should ever feel like their actual skill
is ever going to be replaced. And also, we need it. We're like very physical
tactile beings. And playing instruments is like,
you know, it's a it's a it's an expression. And if anyone feels called to express
themselves that way, then they should. And just to be aware that there is this
potential of, you know, having a bunch of recordings as they get better as
musicians, they could use that as training data. Now we're going through this
interesting era where like artists like Holly Herndon and Grimes are actually
incentivizing people to use their AI generated audio of their voice.
And so I think this is a good step forward because this is how we shift a
narrative and also just like bring more people into the narrative of the the
inspirational side of this technology and the capabilities
♪ Anything I hear ♪ ♪ Something ♪ ♪ I stand here ♪ ♪ Into the pit ♪ ♪ We act
here ♪ ♪ Silence ♪ ♪ We hear ♪ ♪ Reaching ♪ ♪ Against me ♪ ♪ Against me ♪ ♪
Love ♪
(upbeat music)
(upbeat music)
Sittin' on a lie, mmm This
is what it feels like Sittin' on a lie,
mmm
♪ Hey, this is what it feels like ♪ ♪ ♪ Sitting on a wall ♪ ♪ ♪ Silence,
be here ♪ ♪ Me, too ♪ ♪ An artist ♪
"Wire" from the album "Wire" from Portrait XO. I discussed with her also the current
fear of people that AI will take over the art world. And I guess Portrait,
and I share the same opinion there, no, it will not destroy art at all, it will
change it, it will challenge it, so exactly that what always happens when new
technologies take over. I didn't know there were so many misconceptions about AI,
like this misconception that if, you know, you're creative and you're using AI, you
want things to be automated and everything to replace you. And, and it's really odd
because it's like, well, you know, I have so many artist friends around me and we
have these ongoing conversations of supporting each other when we're having these
creative dry spells and stuff. And I don't know a single artist that I personally
know who wants their creativity to be replaced by anything. And if anything,
I think most artists really wanna their own unique style they want to have their
stamp on something that really feels authentic to their most authentic expression of
whatever it is that they're trying to do. We hear another song from that process of
creating plastic skin and I quote Portrait. So here it was written in the
perspective of an AI robot singing to its human creator. The human creates the robot
to fulfill all its desires and needs of emotional and intellectual connection. The
chorus resolves to "Is it the greed? I wish I knew how to love you, but you take
it with no consent and I don't know how to contend. I'm just plastic skin." The
robot basically points out that based on the data analysis of the erratic and
unpredictable demands of human emotional needs, love proves to be inconclusive.
(upbeat
music)
♪ I'm not your hero ♪ ♪ I'm just plastic skin ♪ ♪ With a metal brain ♪ ♪ Is it
too much, too much for you to take ♪ ♪ Plastic smelting ♪ ♪ And when I'm talking
how it ♪ ♪ Oh, in ♪ (audience laughing)
I
wish I knew how to love you,
but you take it with no consent, and I don't know how to content I'm not open,
I'm just happy
Put
your savior
I'm just plastic skin
Rollercoats to clouds each under Or wildfire as they rumble Can we be mad at it?
I wish I knew how to love you,
but you take it with no consent and I don't know how to consent Well,
I'll bring on just practice again [Music]
Plastic Plastic skin from Portrait XO. What I really like about the album wire is
that Portrait had different approaches for each song. So she wrote music with the AI
generated audiophiles each time differently. One good example of this is Vessel,
a co -production with a musician Reza. Vessel is one of my favorite songs because I
got to work with Ramon Rezar and he's such a beautiful songwriter and his voice is
just so so stunning. I'm trying to remember how we met. Ah, yes, okay,
we met when I was getting my in ears made and and he was there and We exchanged
information. It was during COVID. So there wasn't like, you know, people weren't
meeting random people. And so that was like the first time I met, you know, another
artist in like two years or whatever. And so we swapped details.
I listened to his music. I like, I fell in love with his voice. And I have, as
you can tell, I have a thing for Um, I get goosebumps when I hear just like
voices that captivate me, you know, like Sia and Adele, they like, you know,
these are the kind of voices that just like creep into my soul and they get under
my skin, you know, in the best, in the best ways possible. Sometimes people are
like surprised I even bring up like Sia and Adele. Um, but Yeah,
so I fell in love with his voice. We, so AI song contest happened, I think that
was 2021. And that was just after we had met and I was sharing at the time with
musicians about like AI workflows and stuff. And I invited him to ride a track with
me. And so that was our first time getting Um, and it came together so fast,
um, and my approach through to this song that I wanted to achieve was to use my
vocals and his vocals as one data set for training and, um,
I wanted to see if I could find some interesting textural or tonal outputs from
that. Um, and it was really fun fun because Imogen Heap gave our song a shoutout
because she really loved our approach to the experimental side of like,
of trying to merge two voices. I think I underestimated how long it would take to
train this data on this Mel Speck VAE model that a good friend Moises Horta
Valenzuela And, um, I think we trained for like a week and,
um, I think you, I needed to go like another week further or something. But anyway,
we, we worked within the time constraints that we had, um, and just, I just curated
a bunch of sound bites that I thought were the most interesting. And then we
started our session. And I think that's when it's just like so much more of
everything that I had been working and the benefits that I saw start to really be
validated because it was suddenly not me alone in my studio thinking like,
wow, this is really exciting. It was amazing to like bring it into someone else's
world who hadn't been introduced to AI and to see them have a very similar and
just like further validated that, you know, and he also loved the idea of using AI
as like a tool, as an assistant, as a co -creator. And so,
yeah, we put that song down together and like one session, everything came really
quickly. And then, I mean, we obviously had a couple more sessions to clean it up,
But we wrote the lyrics together, and he started the melodies.
He's just such a beautiful writer. Like, he just get him down to hit record and
just melodically. He just had-- it was like his first take. And it was like, oh,
yeah, that's done. So that was really exciting. And I guess for me,
like I really love working with other singers because singing for me didn't actually
come natural.
For the longest time when I was living in LA, I had so many dreams of like
working with other singers that I could write for. So I did entertain the idea of
being a ghost writer for a while.
And I guess life decided to show me different, different journey and so I think
it's something that I still, I'm like always open to and now I am actually in the
process of, of talking to more singers and introduce them to this specific AI model
and this approach because I would love to, to work um other incredible artists like
erica bedou would be so amazing
You know how.
How.
How.
How.
How. How. How. How. How.
How. How. How.
[구독과 좋아요는 필수!]
The
flowers of the Lotus Container Sea vessel shape like an inverted conor bell, which
are very holy symbols for all peoples, and representative male and female.
[MUSIC]
(upbeat music)
Vessel, while I was chatting with PortraitXO we were also talking about other
musicians who use AI to create music. One name that she and I dropped simultaneously
was ReepsOne, an English beatboxer, composer and new media artist. We were both blown
away by his second self -video from 2019. I remember when I saw it the first time
that I had goosebumps. I have never seen a voice AI so human and vivid before.
Portrait will tell you more about that. He is the reason why I actually chose my
voice as a data set to begin with. And it's so funny because I met and Reeps won,
and two different occasions through Music Tech Fest, which is this festival that used
to travel around Europe, and they brought musicians and hackers and scientists
together, and it was like a music hackathon. So we'd have a very short space of
time to develop stuff, like whether it's a prototype for an instrument idea or a
performance. It was just like really But it's so much fun and so that's how I met
CJ and I met Reeps One in Stockholm when we were there for Music Tech Fest.
And so yeah, so they collaborated first before me and CJ did.
And so let's just like rewinding quickly back to 2015 when I first got introduced
and then introduced to AI and then Harry Reeps won,
collaborated with CJ for his residency at Nokia Bell Labs,
and he put out this amazing video called Second Self. I think this was 2018.
This was, and still is, the most incredible performance that exists out there that
truly exemplifies human -machine collaboration with AI because the second self -video is
amazing
Exit like representation of human -machine collaboration and also the tension of it
because you witness Him in this anechoic chamber and he's beatbox battling against
his AI Self and the way the AI self is visualized I mean,
he's just he's also one of my favorite artists and one of my my best friends that
I'm just so honored to have in my life because he's just so incredibly inspiring
and and he really pushes these these grand ideas in such a beautiful way and like
in such a way that also helps us understand as well in in a very explicit way.
So yeah, I I have to remember to reference him as much as possible, because he him
and CJ basically pioneered, I would say, AI and music before anything happened.
And so, yeah, I think it's important to remember, you know, like these,
the first innovators as well, Um, that helped me understand,
like, what does it even mean to pick a data set for trading?
Reaps one with second self. That video created so many emotions I saw it for the
first time, so you should see it too, because just listening to it is different.
Please watch that video. It is spectacular how the AI is presented and interacting
with Reeps. You get the feeling he is singing to a lively thing with a heartbeat
that responds to his beatboxing. I still do not understand why that video has not
millions of views, maybe one say, that will happen when in the future we look back
at this video as a milestone in the AI and music history. But watch it for
yourself. I will link the video and all the other important music in the notes of
this episode. Also the link to the page where you can buy the album from
PortraitXO. So at the moment it's on 12x12 NFT marketplace and you don't need to be
into crypto to support, um, I picked 12 by 12 because you can use just like your
normal credit card. Um, but I guess like this was the, this whole album has been
like a very big research project, everything from the production of it to the
execution. And I guess when I was thinking about releasing the Um,
I was going through this weird like period of trying to understand is like the
importance of copyright. Um, I won't go into this in too much detail, but like I
had a short film idea based on plastic skin, the last, um, one of the last songs
on the album. And I tried to get a copy written, um, via US copyright office.
And I mentioned that there were like AI -generated images that I wanted to use that
integrated into the storytelling, but everything was written by me. And then they
sent all these weird questions back and some of them were so irrelevant, but
basically like bottom line was like anything AI generated is not copyrightable was
like the end message to that. And at the same time I was exploring blockchain, but
I was really weirded out at the time, or like, I think I was still trying to feel
out like this whole NFT thing, you know? 'Cause like, it has a weird connotation
even so now, like some people are really not into it. You say the word NFT, and
it can be triggering for a lot of people. And for me, it was more of an
exploration of, well, how can I give people this album in a way that feels intimate
and also explores the uses of technology that offers us as artists to experiment and
express ourselves more intricately. And so I've just been, I guess,
yeah, exploring that by giving people these digital collectibles. And I guess because
I've been able to break down my process in this way by incentivizing people with
these like little bite -sized things, like small little audio visuals and lyric art
and all of that, it helped me to tell my story in different phases.
So everyone who collects all of the NFTs get the vinyl,
which I just found out how I received 200 of them in the office at 12x12,
so this coming week I will be sitting with them to discuss how we want to execute
everything. And this is my first ever vinyl, so I'm really excited.
I don't know if it's going to be on WEB2 yet, I've been playing around with this
idea. I'm loving the idea and I generated voice finally ends up on analog vinyl and
maybe it is also a symbol a metaphor of things merging together the old and the
new I guess that is the way it should be and always will be and was in music we
are coming to an end here in the next episode we will dive into some dataset
questions concerning music because metadata for music to train models is way more
than data sets with images. I mentioned that before. Why is that? More next time.
This was episode 3 of the Iliac Suite. Thanks to PortraitXO for talking to me.
I hope you enjoyed it. Thanks for listening, humans. Take care and behave.
♪ Never leave ♪
♪ So good for you ♪