00:00:00
just in case you missed it last week
00:00:02
Aphex Twin released a sample mashing app
00:00:04
called sample brain I made a video as
00:00:07
soon as it came out you can check it out
00:00:08
over here audio internet went pretty
00:00:10
crazy about this news some thought that
00:00:13
this was genius until that was AFX Queen
00:00:15
being a troll as usual which by the way
00:00:18
is entirely understandable coming from
00:00:20
the guy who made the track out of phone
00:00:22
pranks or that print is Portrait into
00:00:26
the spectrogram of his own music but I
00:00:29
dug a little bit deeper in the app so I
00:00:32
wanted to make another video about it a
00:00:33
pretty interesting thing that I learned
00:00:35
about the app is that this was
00:00:37
originally inspired by the technology
00:00:39
behind Shazam coding the original blog
00:00:42
article which by the way is really
00:00:43
interesting I'm gonna link it down below
00:00:45
please have a look this idea came about
00:00:47
a long time ago not sure exactly when
00:00:49
2002-ish but when MP3s started to become
00:00:52
a thing when for the first time there
00:00:54
were a ton of them sitting on my hard
00:00:56
drive at the brilliant Shazam had
00:00:58
recently launched started playing
00:00:59
thinking all this music sitting here
00:01:02
maybe can be used for something else
00:01:03
other than just playing or DJing I had
00:01:07
originally contacted the founders of
00:01:08
Shazam to discuss further creative uses
00:01:10
of their genius idea but they were busy
00:01:13
making an automatic DJ program I still
00:01:15
think Shazam could be repurposed for
00:01:17
something incredible but in the meantime
00:01:19
we have sample brain I think this is
00:01:21
really cool if this was a prank and I'm
00:01:24
not excluding that it was it was a very
00:01:26
elaborate one let's get started so first
00:01:29
of all I think the team behind simple
00:01:31
brain realized that most people are not
00:01:34
used to dealing with repositories and
00:01:36
they relabel things in the original Link
00:01:39
in a bit of a clearer way and the git
00:01:42
lab actually changed a lot already in
00:01:44
the last week so let's have a proper
00:01:46
look together now you find what was
00:01:48
before called the binaries under a more
00:01:52
clear download and here you find the
00:01:54
installers for both Windows and Mac
00:01:57
previously there was only one installer
00:02:00
for Intel and then I added a second
00:02:03
installer for M1 now there is a single
00:02:05
installer that works for both and I also
00:02:08
see a old broken spurious binaries so if
00:02:12
you had one of those and you had no
00:02:14
success in installing the app please try
00:02:17
the new one because they clearly updated
00:02:19
something there were a bunch of people
00:02:21
that were supporting having no sound and
00:02:24
no managing to install the app I've also
00:02:27
looked at the issues tab here in the
00:02:30
repository and if you still have in
00:02:32
trouble please submit an issue I'm sure
00:02:34
that this will help the team making the
00:02:36
app better
00:02:37
and if you're on Mac as it's very
00:02:39
typical with non-commercial programs
00:02:41
remember that you need to tell your Mac
00:02:43
that it's actually okay to install
00:02:46
software from this developer and if you
00:02:48
don't know it you do this by going into
00:02:50
system preferences clicking on security
00:02:53
and privacy and when you try to install
00:02:55
it you'll get a message over here and
00:02:58
you just need to click on OK and finally
00:03:00
do keep in mind that this is pretty much
00:03:02
better software one of the questions
00:03:04
that I saw in my comment section and
00:03:06
also Elsewhere on the Internet is
00:03:08
whether this app is neural network based
00:03:10
or not I'm not an expert and I don't
00:03:14
know very much about neural networks and
00:03:16
AI so I can't answer that but I think
00:03:19
that there are some pretty big clues
00:03:21
that disease at least told to be
00:03:23
inserted into a neural network first of
00:03:26
all one of the algorithms is called a
00:03:29
synaptic over here and then there is
00:03:31
this net tab which is really exciting to
00:03:34
me this allows you to control multiple
00:03:37
instances are sample brain over the
00:03:39
network all running their own brain
00:03:42
simultaneously this feature has not been
00:03:44
tested well it's basically telling you
00:03:46
that you can put multiple computers with
00:03:49
several instances of sample brain and
00:03:53
make a neural network with them but I
00:03:56
would say that Clues are even in the
00:03:57
actual description of the app sample
00:04:00
brain chop samples up into a brain of
00:04:02
interconnected small sections called
00:04:05
blocks which are connected into a
00:04:07
network by similarity it processes a
00:04:10
Target sample chopping it up into blocks
00:04:13
in the same way and tries to match each
00:04:15
block with one in its brain to play in
00:04:17
real time but as I said definitely not
00:04:20
an expert I know very little about this
00:04:22
kind of thing so do let me know in the
00:04:24
comments if you know something more okay
00:04:26
let's bring some samples in so I got
00:04:29
some slightly different stuff compared
00:04:30
to last time I have again some cello
00:04:33
samples
00:04:37
foreign
00:04:38
[Music]
00:04:40
booms Some hi-hat Loops
00:04:44
some pretty cool cake loops
00:04:50
I have more trucks of mine that I want
00:04:53
to use as targets
00:05:00
but also like this piano one I was
00:05:02
really curious to see what happens when
00:05:04
the blocks try to match a piano piano
00:05:06
solo track
00:05:09
foreign
00:05:10
and I got some peaked piano samples
00:05:18
I have some post rock guitar samples
00:05:23
I got some snare loops
00:05:29
I got some vocals
00:05:36
a bunch of stuff let's drop all this
00:05:38
stuff in and let's see what happens
00:05:43
I wish there was a drag and drop version
00:05:45
by the way or that at least you could
00:05:47
put more than one sample at once but
00:05:51
I understand
00:05:54
and then let's load one of the target
00:05:56
let's start with the let's start with
00:05:58
the piano track actually
00:06:00
I think could be cool
00:06:02
by the way I couldn't find a way to
00:06:04
change the dark mode for this one I
00:06:06
normally like apps in dark mode but this
00:06:09
one is not very readable because of
00:06:11
these white over bright gray
00:06:14
highlighting I think it just takes it
00:06:17
from the system so it would have to
00:06:19
change the preferences on my entire
00:06:21
system but it doesn't matter it is
00:06:23
really not a big deal let's leave all
00:06:24
the samples checked so it might be a bit
00:06:27
messy and let's generate the brain
00:06:31
processing samples building synapses and
00:06:35
then let's generate the blocks there you
00:06:38
go and now we should be able to press
00:06:39
play
00:06:40
we have sound so this is my brain and
00:06:44
this is my target
00:06:53
very cool let's start looking at some of
00:06:55
the parameters the first parameter that
00:06:57
is worth having a look at is this first
00:06:59
one which basically decides on the model
00:07:02
that is being used to create the
00:07:05
analysis of the spectrum so ffd is a
00:07:09
very common thing in a DSP processing
00:07:11
and it stands for phosphorier transform
00:07:14
here it calls a rule frequency analysis
00:07:17
on the manual but I'm pretty sure that
00:07:18
that's what they're using and mfcc is
00:07:21
something that I haven't heard of is Mel
00:07:23
frequency sexual coefficients I'm sure
00:07:26
if we give it a Google we will return
00:07:30
some meaningful results but the
00:07:33
interesting thing here is that you can
00:07:35
blend between these two models and
00:07:37
putting one to zero percent or one
00:07:39
hundred percent bypasses the other which
00:07:42
makes it a bit more CPU efficient which
00:07:45
is quite helpful because one of the most
00:07:47
common issues that I found mentioned
00:07:49
online is that this thing is a CPU hog I
00:07:53
have a pretty powerful system so I
00:07:55
haven't really noticed this all that
00:07:57
much about I'm sure it's true because it
00:08:00
is indeed doing some pretty complicated
00:08:02
stuff in the background
00:08:05
[Music]
00:08:17
you can really doubt that by using all
00:08:19
the samples there is a whole bunch of
00:08:21
stuff but it's quite interesting so for
00:08:23
example one thing that I want to do
00:08:25
right away is to for example let's only
00:08:27
use the kick drums to try and recreate
00:08:31
the target
00:08:33
and so we need to regenerate brain and
00:08:35
we should have some pretty different
00:08:37
results obviously you're essentially
00:08:39
changing the ingredients that is using
00:08:42
to attempt a match of the target sound
00:08:50
[Music]
00:09:08
really cool
00:09:13
very interesting here you can sort of
00:09:15
hear the original track through some
00:09:20
sort of processing
00:09:23
like if it's in a stock buffer or
00:09:26
something
00:09:27
just for reference let's have a listen
00:09:29
to the original track which is this one
00:09:33
[Music]
00:09:40
you can really tell that it's using this
00:09:43
as well
00:09:44
you can mix between the chair
00:09:47
[Music]
00:09:49
really cool by the way stereo mode is
00:09:51
something that we can try right away is
00:09:53
essentially generating different streams
00:09:56
for the left and right channel so the
00:09:58
results are stereo but you can also use
00:10:01
them as dual mono
00:10:04
[Music]
00:10:12
really cool
00:10:16
thank you
00:10:18
really interesting moving on the second
00:10:21
option is essentially deciding whether
00:10:24
you're using both frequency and Dynamics
00:10:26
to attempt a match or only the frequency
00:10:33
[Music]
00:10:35
but basically this is using the
00:10:37
normalized blocks
00:10:40
[Music]
00:10:43
so it's not checking for the dynamic
00:10:45
changes in the brain contents to try and
00:10:49
match the target sound if you read in
00:10:51
the manual all these parameters are
00:10:53
really interesting and explained in
00:10:55
details and I love that they included
00:10:57
things like the fft subsection but it
00:11:01
also says not the useful in practice so
00:11:03
far you can really tell that this is an
00:11:06
experimental thing that they've been
00:11:07
working on and they're including
00:11:09
controls that might play a larger role
00:11:13
in the future but I'm just going to
00:11:15
guide you through what I think are the
00:11:17
most useful so far and the things that
00:11:21
are the most impact on the sound novelty
00:11:24
and boredom are definitely an example of
00:11:27
such thing so they work together and
00:11:30
reading the manual you can use novelty
00:11:34
to bias the selection away from
00:11:36
similarity and prioritize similar blocks
00:11:39
we haven't used yet so it changes the
00:11:42
block that is current being used and
00:11:44
boredom increases the speed at which
00:11:47
novelty wears off creating a wider
00:11:49
spread of possible blocks to be used
00:11:51
there are a lot of things that are
00:11:54
somehow similar to the base to the
00:11:57
concept behind granular synthesis let's
00:12:00
have a listen to this
00:12:02
[Music]
00:12:17
another cool thing is the stickiness is
00:12:20
we'll have the effect of elongating
00:12:22
chunks of rain samples that you hear so
00:12:25
it's pretty cool actually
00:12:29
as opposed to
00:12:33
then the third stretch uses basically a
00:12:37
Time scratch
00:12:39
foreign
00:12:40
[Music]
00:12:50
Loop and some of the snares Maybe
00:12:53
let's regenerate the brain
00:12:58
can hear some of the hi-hats and snares
00:13:01
the frequency content changed
00:13:03
dramatically
00:13:08
really cool
00:13:16
then here you have all the different
00:13:18
algorithms we change the sound
00:13:19
dramatically I'm not going to go through
00:13:21
each one of them but please do have a
00:13:23
look at the manual because it's actually
00:13:25
really interesting let's listen to that
00:13:27
so this basic is reversed
00:13:34
is synaptic
00:13:39
[Music]
00:13:45
and they slide
00:13:54
okay I can't really telling synaptic as
00:13:58
light oh yes I can
00:14:00
how really cool
00:14:04
so these uh number of synapses and
00:14:08
synapse light error only work in
00:14:10
synaptic and slide modes so let's have a
00:14:13
listen to those
00:14:15
this increases the number of synapses
00:14:18
being used obviously
00:14:23
synaptic slide error
00:14:29
this is quite interesting in the menu it
00:14:31
says that this controls the acceptable
00:14:32
error to consider a block as close
00:14:35
enough in slide mode so this works in
00:14:38
slide less Elizabeth
00:14:40
okay
00:14:43
there you go let's decrease the novelty
00:14:53
foreign
00:14:57
lots of fun the next section are
00:14:59
parameters that control the way the
00:15:01
target sound is broken up into different
00:15:04
blocks so how about we change Target
00:15:07
let's uh use another track of mine
00:15:11
this is the raw track
00:15:14
[Music]
00:15:19
go through it
00:15:24
[Music]
00:15:30
okay should be fun and let's regenerate
00:15:33
the blocks
00:15:34
[Music]
00:15:36
let's bring you back to basic
00:15:39
[Music]
00:15:47
so fun so block size basically decides
00:15:51
the size of the blocks in the sample
00:15:54
let's see what happens
00:15:59
foreign
00:16:00
[Music]
00:16:08
I hate that it's white on white very
00:16:10
unreadable one thing that might be very
00:16:12
fun is to use different block sizes for
00:16:15
the Target sound for the brain contact
00:16:16
so let's change it let's see what
00:16:18
happens
00:16:19
[Music]
00:16:26
then the block overlap controls the
00:16:29
proportion to overlap the block
00:16:31
generation not sure what that means less
00:16:34
have a listen
00:16:37
okay so these are ratio essentially
00:16:40
let's try and decrease it
00:16:42
0.2
00:16:49
the window shape is also interesting
00:16:52
essentially are a volume envelopes that
00:16:55
are applied to uh the different blocks
00:16:58
uh before the analysis the analysis uh
00:17:01
you can experiment with different ones
00:17:04
let's have a listen
00:17:05
has been the block overlap to 0.80 I
00:17:08
think it was before let's try different
00:17:10
ones
00:17:13
let's listen to the brain only
00:17:16
Bartlett
00:17:19
[Music]
00:17:23
I actually think I might need to
00:17:24
regenerate the blocks because this
00:17:27
happens before the analysis so to hear a
00:17:29
difference I need to regenerate the
00:17:30
blocks
00:17:32
okay very different
00:17:37
last year
00:17:39
flat top
00:17:46
very fun
00:17:47
so I'm actually sure that I had to
00:17:50
change regenerate the blocks also once I
00:17:53
change the block size and the block
00:17:54
overlap so that was a bit dump let's try
00:17:57
put 0.4 and block size to 2000.
00:18:03
and maybe let's change the block size
00:18:05
here to I don't know uh
00:18:09
500 literally putting random numbers in
00:18:12
so let's regenerate the brain
00:18:17
this is taking a long time
00:18:21
wow building the synapses now is taking
00:18:23
uh way longer than it used to uh so I'm
00:18:27
pretty sure that the block size
00:18:28
influences very much the amount of time
00:18:31
it takes to generate the brain which
00:18:33
makes sense
00:18:38
it's done let's also regenerate the
00:18:40
blocks which I expect gonna take a long
00:18:42
time as well actually not that much
00:18:45
let's try it
00:18:48
okay
00:19:04
oh that's why I put 20 000 rather than
00:19:07
two thousand so let's try with two
00:19:08
thousand and let's regenerate the blocks
00:19:12
so it was getting pretty huge uh block
00:19:15
sizes
00:19:20
[Music]
00:19:28
it is pretty cool
00:19:30
the parameters for the brain contents
00:19:32
are identical to the Target sound so I'm
00:19:35
now going to repeat myself but these are
00:19:38
all pretty cool things that we can
00:19:40
experiment with and then let's have a
00:19:43
just a brief look at the mix we already
00:19:45
sort of uh briefly mentioned it the
00:19:49
auto-tune parameter is going to force
00:19:52
the pitch or attempt to force the pitch
00:19:54
of the incoming block from the brain
00:19:56
contents into the block of the target
00:19:59
sound that is trying to match
00:20:01
foreign
00:20:06
parameter is essentially flattening the
00:20:08
dynamic and thus increasing the volume
00:20:13
it is designed to work with the
00:20:14
frequency only
00:20:16
search as we said which makes sense you
00:20:18
have less variations in Dynamics
00:20:20
so it's basically not taking the dynamic
00:20:23
from the actual brain content but it's
00:20:26
generating a Xposed and then as we
00:20:29
mentioned the brain Target is going to
00:20:30
blend between the two
00:20:36
it's quite cool to double check the
00:20:39
pitch
00:20:45
there you go let's leave it all to break
00:20:47
and then as I said the stereo mode
00:20:49
basically runs separate block searches
00:20:52
for the left and right speaker you can
00:20:55
record the samples so let's make some
00:20:58
samples
00:21:01
[Applause]
00:21:05
[Music]
00:21:14
foreign
00:21:23
[Music]
00:21:37
now let's open something like Ableton
00:21:40
and let's do some editing and cleaning
00:21:42
up or actually let's use Cubase first
00:21:44
because I have a quicker way to batch
00:21:46
export my edits once I've done that
00:21:49
let's grab all the left samples
00:21:52
put them here
00:21:54
now let's get all the right samples
00:21:57
let's put them here
00:22:02
[Music]
00:22:06
now that's hard pan boat track or the
00:22:10
left and our right
00:22:16
let's try and make a new macro
00:22:21
select and create cycle markers
00:22:25
okay
00:22:35
and now we can export the audio mix down
00:22:39
the export window we're going to select
00:22:41
multiple export all cycle markers
00:22:45
and in the name I'm going to use the
00:22:47
circular marker name and Sample brain
00:22:50
here's an underscore as a separator okay
00:22:53
export audio
00:22:56
pretty cool now let's import stuff into
00:22:59
the Ableton sampler I mean now we can
00:23:02
really have a lot of fun we have 10
00:23:04
pretty cool samples
00:23:23
lots of fun or we can even import them
00:23:27
as separate audio loops and let the
00:23:30
analyze with the warp
00:23:48
another one
00:23:51
let's try this one
00:23:54
tonight obviously Tempo thing
00:23:56
[Music]
00:24:06
to bring some effects
00:24:09
let's start with the reason address
00:24:12
which I normally love
00:24:15
let's try with Berlin
00:24:18
nice
00:24:29
[Music]
00:24:35
foreign
00:24:38
[Music]
00:24:42
stuff I also want to make a contact
00:24:44
instrument let's try that let's make a
00:24:47
new instrument
00:24:48
so on the mapping editor and let's drag
00:24:51
and drop these samples in let's drop
00:24:54
them in let's actually make one per
00:24:57
pitch
00:24:58
there you go so you have them here long
00:25:00
octave
00:25:01
foreign
00:25:03
[Music]
00:25:10
foreign
00:25:14
[Music]
00:25:18
ly I can obviously make different groups
00:25:21
with them but let's just have a quick
00:25:23
look so one thing that could be cool be
00:25:26
to put them in a time machine pro mode
00:25:29
or maybe beat machine actually let's do
00:25:31
that and we can open the wave editor and
00:25:34
let it do the analysis go on auto and
00:25:37
it's going to select a number of slices
00:25:43
to go pretty down for this one let's try
00:25:46
it
00:25:51
I'm gonna go do one by one
00:25:56
but it should be pretty cool and now if
00:25:59
I change the speed to Zone becomes Tempo
00:26:02
synced let's slow it down a lot
00:26:12
[Music]
00:26:15
put it to I don't know 180 BPM or
00:26:18
something like that
00:26:20
foreign
00:26:21
[Music]
00:26:45
[Music]
00:26:53
I love this one
00:26:58
let's tweak this one a little bit
00:27:01
[Music]
00:27:04
there's glitch in a box really cool
00:27:07
let's Loop it
00:27:15
so let's put that the first one is only
00:27:18
gonna play two times and then it goes on
00:27:21
the second one is gonna play
00:27:23
three times
00:27:26
foreign
00:27:31
[Music]
00:27:39
really cool obviously you can do a lot
00:27:40
of stuff
00:27:41
and now the modular is calling my name
00:27:50
[Music]
00:27:56
hopefully this was fun in conclusion I
00:27:59
really don't think this was a prank and
00:28:01
I loved making samples with it it's
00:28:04
obviously glitchy as hell but hopefully
00:28:06
it was clear how much fun you can have
00:28:09
with it one of those always appreciated
00:28:11
please subscribe to the channel if you
00:28:13
want to be notified when I put a new
00:28:15
video up thank you so much for watching
00:28:16
I'll see you next time
