Did AI Just End Music? (Now it’s Personal) ft. Rick Beato

this video was brought to you by

brilliant hi welcome to another episode

of Cold Fusion The Following song is

100% generated by AI

[Music]

abusion where

curiosity ignites and Minds take

[Music]

life a better future

can help us make a better future

everything you hear is all artificially

generated by AI generated I mean the

lyrics which was chat gbt the Rhythm the

vocals to get this output all I did was

type in a text prompt of what I wanted

there are a few revisions but still the

result is honestly insane and if you

think that's a one-off here's another

example I wanted a classical piece so I

type this

[Music]

how about some guitar driven

stuff

so you

cra a wild ride awesome

[Music]

[Applause]

R&B our souls please remember me

[Music]

awesome UK

[Music]

[Applause]

garage all very impressive even the

album covers were made using Ai and

that's just to let you know what we're

dealing with

here as some of you know I've been a

musician for the better part of 20 years

and a producer for a decade if you've

watched a cold fusion video before

you've heard some of my music so being a

technology Enthusiast as well a do have

mixed feelings about this AI has finally

come to something very close to my heart

we've seen AI voice swapping that could

get artists to perform songs in

different genres but this is something

totally different with this latest

generation of AI music it's not just the

vocals but the backing music which is

artificial which in my view is much more

difficult for AI to do coherently what

you just heard previously was from a

platform called udio and it was created

by a group of X deepmind Engineers it's

a watershed moment for AI music and

they're not the only ones a few months

ago there was a similar application

called sunno AI in a recent video I

looked at the AI deception and it was

about how some companies are lying and

over promising the capabilities of AI

but on the flip side I said that there's

also some companies that are obfuscating

the true extent of AI replacing jobs

today's episode falls in the second

category first was images graphic design

writing videos and now finally music

it's clear that AI music systems will

will eventually impact those in the

music industry but how in this episode

we explore AI music apps what they're

all about how it works its criticisms

and how AI could disrupt the music

industry I also had the opportunity to

speak to the amazing musician and

YouTuber Rick Bardo to get his thoughts

on this and I also spoke to Taran

Southern who's considered to be one of

the first artists to commercially

release music using AI all right so on

the count of three let's

go you are watching to Fusion

TV quote creating music is an inherently

complex process but we're streamlining

that making it more accessible and

intuitive than ever before end quote

that's from son's co-founder Mikey shaan

and it just about summarizes what these

AI companies are trying to achieve in

fact the premise of sunno and udio is

basically the same now anybody can

create music even though they have no

prior musical knowledge just type a text

prompt and Away you go users can provide

subjects instruments feelings and or

their custom lyrics and in just a minute

a track is ready to play 30 Seconds for

udio and 2 minutes for sunno both

platforms can extend tracks and provide

different variations and both are free

to use right now some say that Yo's

first version is already pretty good

better than Sun's version 3 I've used

both of them to gain Insight but I ended

up liking y's output more personally

because it was cleaner and possessed a

better understanding of less typical

genres so even though the examples I've

shown you and will show you are through

using yio keep in mind that a lot of the

content in this video also applies to

sunno sunno version 3 is now available

to everyone including those on the free

plan it is the best AI music generator

by far and it just got even better with

V3 a completely free AI music generator

launched called udio people have been

saying this is the chat GPT moment for

AI music or calling it The Sora of Music

a lot of the time people massively

overhype new AI launches that's how

everyone gets clicks but there's some

truth to this one to understand just how

revolutionary this Tech is it helps to

know how modern music is made before AI

the first step is to come up with a

guitar refi I made this one a few years

ago when marking around on my acoustic

guitar but for the final track I used a

Gretch electric guitar with a delay in

Reverb pedal and a distortion pedal I

was going for a calm 34 Walt time

signature with a post Rock feel

[Music]

after I recorded this I hopped into

Ableton and built some Bas and drums and

atmosphere around the guitar ref even

for this process every step involves

many substeps including eqing

compressing adding effects and just

listening to the components in isolation

and

[Music]

together as the song is being created

you'll definitely come across some

production problems where frequencies

Clash solving these issues are just a

part of the nature of mixing sounds

together finally I added some vocals and

then arranged the

track so in total a finished song can

take days or even weeks to complete and

sometimes the songs just don't work out

so you have to try again but with all of

that effort exploring the Sonic stage

and pulling a song out of the ether is

half the fun but now in contrast with AI

all people have to do is type in some

words and then get a song I hope you can

see the difference clearly now here's

some observations that I found with yudo

it includes production techniques like

side chaining tape effects Reverb and

delay on vocals in appropriate areas in

some outputs the vocals seem extremely

real there's also harmonies in there and

even for electric guitar it mimics the

sounds of different pickups all just so

insane some weaknesses include sometimes

it messes up big time and I'll show a

couple of examples of

[Music]

that it's very limited in flexibility

once you get an output you can't really

change it there's low Fidelity on some

tracks and it has some weaknesses in

some genres such as UK jungle okay so

what's the story behind these

applications sunno is a venture-backed

company founded in late 2023 all four of

the founders previously worked at Keno

an aite Tech startup for financial data

which was later acquired by S&P Global

sunno even partnered with Microsoft

co-pilot and is up to version 3 like

sunno udio was founded in late 2023 but

the company only recently went out of

stealth mode and made their application

public a couple of weeks ago it was

founded by an ensemble of former

researchers from Google Deep Mind and

has the financial backing of popular

Silicon Valley venture capital firm and

seen hwart and also from musicians like

common and Will i Am as you've seen well

heard the output is a rich combination

of all sorts of instruments so the

training data must have been significant

and that leads us to the next section

how do these applications work well

we'll dive in in a bit but first let's

rewind the tape a bit this isn't

strictly the first time this has been

tried making music with the aid of

computers goes back to 1957 lejaren

Hiller and Leonard isacson would compose

a piece called the iliac Suite this

piece of music is often considered to be

the first created with the aid of a

computer there were notable efforts in

the 1980s to point computers into a new

Direction generative music in 1984

scientist and composer David cope

created Emmy which stands for

experiments in musical intelligence it

was an interactive software tool that

could generate music in the styles of

different composers from short pieces to

entire operas and I said you know what

if I could create a

program that would create B carals and

cope and Stravinsky and Mozart and

anybody and the only way I could think

of doing that was to create a program

that was built around a database of

Music let's say I had a database of Bach

Corrals all 371 of them and he had a

little program that was sat on top of

this smallest program I could make that

would analyze that music and then create

a new Bo Corral that wasn't one of the B

Corrals that was in the database but

sounded like them all and some of the

outputs had been used in commercial

settings over the years there's a very

interesting story regarding Emmy and

I'll get into that later in the episode

so stick around for that and then there

was the computer accompaniment system in

1985 it used algorithms to generate a

complimentary music track based on a

user's live input the computer

accompanyment system is based on a very

robust pattern matcher that compares the

notes of the live performance as I play

them to the notes that are stored in the

score

even David Bowie developed a tool in the

'90s it was called the verbasizer a

digital approach to lyrical writing well

that's all well and good you might be

thinking but what about the modern stuff

well as you probably know from 2012 the

world switched from hard-coded narrow

algorithms to neural networks and with

that the modern AI boom had officially

started but the the question has to be

asked how are neuron Nets being used in

music Generation Well in 2016 Google's

Project magenta turned a few heads when

they released an AI generated piano

piece made by Deep learning algorithms

ultimately the next big step was to

generate music just based on a simple

text query in natural human language

Google's music LM did just that and

honestly it was impressive for the time

but far from perfect this was followed

by open ai's jukebox stable audio

adobe's music generation gen and

YouTube's dream track now these were all

Valiant efforts but they all had the

same problem they were ultimately fairly

limited they worked to a degree but more

often than not the composition sounded

rigid and just not very human to put it

bluntly but even with such limitations

there were some trying to push the

boundaries Taran Southern is often

considered to be one of the first

musicians to use modern AI to release

commercial music and this was all the

way back in 2017 if you can believe it I

asked her all about it I was having

trouble finding really high quality

music for the documentary that was

relatively inexpensive and so I was

looking for alternate options and I came

up upon this article in the New York

Times that was looking at artificial

intelligence as a way to compose music

and I thought well that's really

interesting so I started experimenting

with it was blown away by what was

possible you know this was even before

llms hit the scene and I think where we

were at with music creation and AI back

in in 2017 was actually pretty far along

um and so I was so excited by what I was

hearing that I just decided to create an

entire experiment out of out of the

project and made an album using I think

four different AI Technologies keep in

mind that what Taran did back in 2017

was quite different to what's going on

today there was still a lot of work that

had to go into the compositions but now

anyone can type in some text and get a

decent output so how does it all work

what's the tech behind it

like most modern generative AI

applications for example llms these

systems use vast amounts of data to

understand patterns and then create an

output based on the user's input for

chat GPT the output is text and for

these AIS its original songs and lyrics

we'll touch on the ethical concerns a

bit later but you can probably realize

something a large language model like

chat GPT for example generates text by

predicting the next word in a text

sequence but composing music is

significantly more complex due to a lot

of variables instrument tone Tempo

Rhythm sound design choices volume of

different components compression and EQ

are just some of the variables that has

to be considered in addition the system

must understand what the user wants and

how that relates to genres and

particular sounds in a coherent yet

pleasing way not easy by any means but

what about the audio synthesis itself

many point to audio diffusion as the

secret Source behind generative music in

simple terms it's is the process of

adding and removing noise from a signal

until you get the desired output we

first saw similar methods used in images

where image diffusion starts off with

random noise which is then refined to a

final image based on the interpretation

of the prompt audio diffusion Works in a

similar way this is a very high level

breakdown just to save time but if you

want to get into the depths of the

technicalities I'll leave a link to a

few articles in the source document in

the video description as always

if you've been following the generative

AI space for the last 2 years you

already knew this part of the video was

coming open AI CTO Mira morati was under

public fire following an interview about

the training data of their video

generation tool Sora when asked if the

impressive video output from Sora was

the result of data trained on videos

from YouTube Facebook and Instagram she

said that she quote was not sure about

that to many online the answer was

suspicious with only a few weeks since

the launch of yudo and Soo version 3

there's a question that has to be asked

were these systems trained on

copyrighted material one user set out to

investigate they ran their test by

entering in similar lyrics to the song

Dancing Queen by EBA in the prompt they

had no direct mention of the band or the

song but the outputs were very close to

the original including the basic Melody

Rhythm and even the Cadence of the

vocals now even human musicians are

continuously accused and taken to court

for a lot less as the user later point

points out it's not impossible to

achieve these results without the

original song being present in the data

set but the similarity is striking this

result was also solidified by other

experiments carried out by the user for

obvious reasons I can't play any of the

copyrighted music here to show you the

comparisons but as always I'll link the

original article in the show notes but

here's the thing it seems like the

founders already knew that this was

coming and they accepted it as part and

parcel of the new AI startup culture

anonio Rodriguez one of sunno ai's

investors told Rolling Stone that he was

aware of potential lawsuits from record

labels but he still decided to invest

into the company sunno AI has stated

that it's having regular conversations

with major record labels and is

committed to respecting the work of

artists if that's actually true is

questionable especially when a few weeks

ago over 200 plus artists wrote an open

letter to AI developers and tech

companies to quote cease the use of AI

to infringe upon and devalue the rights