From Pause Tapes to Neural Synthesis: AI and the Future of One-Shot Sample Design with Max Shafer of Just 4 Noise

WEBVTT
00:00:00.119 --> 00:00:01.524
It's like my drum machine broke.
00:00:01.524 --> 00:00:03.009
So I founded a whole company.
00:00:03.009 --> 00:00:05.708
I made a whole tool to make drum samples from.
00:00:05.900 --> 00:00:07.727
Realistically, we would have probably done it eventually, right?
00:00:17.420 --> 00:00:21.111
I'm joined today by none other than the Max Shafer.
00:00:21.111 --> 00:00:23.868
He is CEO and co-founder at Just 4 Noise.
00:00:23.868 --> 00:00:25.205
Max, great to have you today.
00:00:25.379 --> 00:00:27.204
Tom, thank you very much for having me as well.
00:00:27.204 --> 00:00:35.667
I'm honestly quite a big fan of your show so far, as you know, you've had some of my friends on as well and really enjoy what you're doing and think what you're doing is super cool for the community.
00:00:35.667 --> 00:00:38.159
So when you asked me to join, I was super happy.
00:00:38.399 --> 00:00:38.761
Awesome.
00:00:38.761 --> 00:00:43.070
Well, I'm looking forward to learning a bit more about your company and jumping into your track selection.
00:00:43.070 --> 00:00:45.701
So on that point, you run Just 4 Noise.
00:00:45.701 --> 00:00:46.844
Pretty new startup.
00:00:46.844 --> 00:00:49.311
Tell us a bit more about what it is and who it's for.
00:00:49.600 --> 00:01:04.992
Yeah, so we're about four months old now, so we're quite young, and basically what we're building is a suite of sample generation tools which essentially helps music producer cut the amount of time it takes to go from idea in their head to a sample that they can then use directly in their tracks.
00:01:04.992 --> 00:01:15.403
Our ethos, I would say, is not only this idea of time saving for music producers, but also the ability for music producers to create sounds that essentially don't exist in the universe right.
00:01:15.403 --> 00:01:37.506
So we've built a handful of machine learning models trained around an immense amount of high quality sound data, and through this, what we call neural synthesis rather than like a typical synthesizer, we call it a neural synthesizer You're able to generate sounds that are not only high quality and match the description that you give it, but also don't match any of the data that the models are trained on one for one.
00:01:37.506 --> 00:01:44.326
So every time you prompt the generation of a new sound, it's kind of unique in that sense, which is also a super interesting part of the of the product.
00:01:44.346 --> 00:01:53.367
Okay, so we'll get to the neural synthesis bit in a second, because that's interesting if we think about how the current landscape looks for producers.
00:01:53.367 --> 00:01:58.043
You kind of got the classic sample libraries on one side so you can go to Splice.
00:01:58.043 --> 00:01:59.388
You can get the sounds you want.
00:01:59.388 --> 00:02:03.909
Then you have the more generative AI tools like Udio and Suno, etc.
00:02:03.909 --> 00:02:05.897
Where do you sit within this?
00:02:05.897 --> 00:02:09.108
Because there's a lot of noise right now about this and you are Just 4 Noise.
00:02:09.108 --> 00:02:09.108
.
00:02:09.108 --> 00:02:09.108
.
00:02:10.050 --> 00:02:10.572
That was a good one.
00:02:11.061 --> 00:02:14.050
So explain a bit more where you sit within that landscape.
00:02:14.400 --> 00:02:15.486
Yeah, it's really interesting, right?
00:02:15.486 --> 00:02:23.305
I mean, when we tell people, yeah, we're building generative AI music tools, people automatically go oh, you're like Suno and Udio and everybody has their own.
00:02:23.870 --> 00:02:25.060
Just another one of those right.
00:02:25.120 --> 00:02:29.661
Exactly, and everybody has their super strong opinions about those companies, and so do we, of course.
00:02:29.661 --> 00:02:34.194
But we kind of see ourselves primarily as a music producer tool.
00:02:34.194 --> 00:02:49.413
Right, we're a neural synthesizer that's what we like calling ourselves because we're just another synth that a music producer can leverage in their workflow, and while a music producer is able to generate the sounds via AI, I would say that's kind of not really our boasting point.
00:02:49.413 --> 00:02:58.548
All we're trying to do is take that search that you would typically do in Splice or other loop platforms and sample libraries and turn it into a generation rather than a search.
00:02:58.548 --> 00:03:11.548
And that was I mean I'll get into the lore a little bit later of how we thought of this idea and stuff like that, but that's kind of the genesis, like we as music producers and also all of our friends, always complain about not only sample searching but also sample management.
00:03:11.788 --> 00:03:27.050
Right, I mean, there's so much data in the world If you look at Splice alone, we're talking eight, nine figures of data, right, of just one shot samples alone and we thought solving that search problem is nearly impossible because it takes tagging the entire world internet of sounds.
00:03:27.050 --> 00:03:38.402
But we realized that generation instead via neurosynthesis is a very, very good solution to get from that idea of a search term to a high quality sample that you can actually go ahead and use.
00:03:38.402 --> 00:03:41.711
We are also only focusing on one-shot samples to begin.
00:03:41.711 --> 00:03:44.748
The big ones in the AI space are doing song generation from text.
00:03:44.748 --> 00:04:00.451
A lot of others are doing loop generation or small format generation, but we're focusing only on one-shot samples, which is really interesting for us because one it kind of again pushes us in this direction of developer tools rather than creativity generation or something like that.
00:04:00.451 --> 00:04:10.641
But we're also able to gather data more easily because of that and also gather high quality data and produce high quality sounds in the end, because we're able to train on higher quality data at higher rates.
00:04:11.203 --> 00:04:11.524
Got it.
00:04:11.524 --> 00:04:14.570
I mean, for me this almost flips the idea inside out.
00:04:14.570 --> 00:04:20.699
So when I first came across you, I'm like, okay, this seems like just another version of a sample library.
00:04:20.699 --> 00:04:22.928
That kind of maybe gets me there quicker.
00:04:22.928 --> 00:04:33.192
But this idea of the synthesis, which, when I was looking through the product in more detail, I did actually start to think, okay, this is kind of like a synth, you almost start with a base and you can tweak stuff.
00:04:33.654 --> 00:04:37.305
Tell us a bit more about this idea of the neural synthesizer or neural synthesis.
00:04:37.807 --> 00:04:46.502
Yeah, I mean to be honest, like the only reason we call it that is because, um, the way the machine learning workflow is built is based off of neural networks, right?
00:04:46.502 --> 00:04:53.543
So all it is is we're essentially creating a network of sounds, like a sound world you might want to call it.
00:04:53.543 --> 00:04:53.723
So.
00:04:53.723 --> 00:05:11.963
For example, one of our models is based just on kick drums, and when we train our models from end to end, we're taking around 10,000 sounds, creating this little universe of it, or a neural network, and then from there the synthesis process is just picking a point in this three-dimensional or high-dimensional space and turning it into a sound.
00:05:11.963 --> 00:05:16.987
And that's the process of synthesis via neural networks, and it's a super fascinating process.
00:05:17.569 --> 00:05:23.466
And what's getting really interesting for us right now is we're thinking of this concept of continuous neural synthesis.
00:05:23.466 --> 00:05:42.896
So if I say, hey, give me a dark, clicky 808 drum, and I get a few examples from the model and then I say, oh, this one's really nice, make it darker or make it clickier or make it harder, right, so we're right now looking into this idea of continuously modifying sounds via this neural process over time.
00:05:42.896 --> 00:05:55.237
I think what we're building is going to be the state of art of workflows in the future for all sorts of music production workflows, whether or not it's coming from us or somebody else, who knows, but for sure it'll be the future.
00:05:55.437 --> 00:06:04.649
I mean, you mentioned the word workflow a few times at the end and I think that's the key for me when I'm thinking about, okay, how does this fit into a bigger process of creating a track?
00:06:04.649 --> 00:06:11.410
In your opinion, where does Just 4 Noise come in as a producer, Is it like, okay, I'm starting a track, I need some inspiration, need some sounds?
00:06:11.410 --> 00:06:16.165
Or is it like, okay, I need to get a specific like snare that fits in with what I've got?
00:06:16.165 --> 00:06:17.910
Or is it a bunch of different things?
00:06:17.910 --> 00:06:20.201
Let us know what that could look like practically.
00:06:20.362 --> 00:06:21.223
Really cool question.
00:06:21.223 --> 00:06:30.016
I mean, the way people use samples differ from person to person, genre to genre and also based off of your level as well, whether you're a beginner or advanced.
00:06:30.016 --> 00:06:45.536
The way I would personally use it and the way we map it out in terms of how a producer could implement this into their workflow is as you're working on a track and you begin throwing kind of filler sounds into the track to create a basic arrangement and have a concept for a project you're working on.
00:06:45.536 --> 00:06:49.745
Towards the end, when you get to a point where you say, okay, I really like how this sounds for the most part.
00:06:49.745 --> 00:06:53.689
Now let's kick it up a notch and get better sounds for each of the different elements.
00:06:53.689 --> 00:06:53.911
Right?
00:06:53.911 --> 00:06:58.644
That means putting post-processing effects on some of your voices, modifying some of your drums and so on.
00:06:58.704 --> 00:07:13.129
That's the point where we see people saying, okay, I like this preset that I'm using right now in Ableton, but I want something that sounds more like this Let me go to Just 4 Noise, quickly type that in, see what it gives me, tinker with that a little bit and then bring that into my sampler and begin using it.
00:07:13.129 --> 00:07:19.322
That's kind of the workflow that we see, but it could also work on the front end of the production lifecycle, right?
00:07:19.322 --> 00:07:29.442
You can say I want to start a new project and I want to do hard style, and I've never done hard style before.
00:07:29.442 --> 00:07:34.642
So let me quickly see what Just 4 Noise thinks a hard style kick sounds like or a hard style hi-hat, and let me play around with that and then use that as inspiration for what direction I take the song in.
00:07:34.642 --> 00:07:41.100
That's also possible, but either way, the key is, however, you integrate samples into your workflow.
00:07:41.100 --> 00:07:45.810
The point is we're trying to save the amount of time it takes to find the one that sounds right for your project.
00:07:45.810 --> 00:07:47.374
That's kind of our focus.
00:07:47.800 --> 00:07:48.360
It's super cool.
00:07:48.360 --> 00:07:51.588
I guess, in a way, it's a more advanced search.
00:07:51.588 --> 00:07:54.867
So, like you're saying, you know I want this type of kick.
00:07:54.867 --> 00:08:00.062
You might go to Splice and you search for a genre and you start going through the loops of the different things that are there.
00:08:00.062 --> 00:08:08.274
Or you could go straight to Just 4 Noise or an equivalent and say, okay, give me an 808, like, say, a punchy 808, that's bright, or this or that or the other.
00:08:08.274 --> 00:08:09.925
I think that's a really cool way.
00:08:09.925 --> 00:08:10.749
Or you could do both.
00:08:10.749 --> 00:08:19.355
I guess you could start off with, like, a splice loop and say, you know what, I wish this was a little bit different here and get a one shot to augment that.
00:08:19.540 --> 00:08:20.721
Yeah, it's versatile in that sense.
00:08:20.721 --> 00:08:29.855
And it's interesting when you talk about workflows and we talk about this being the future for workflows and thinking about how much production workflows are changing in general.
00:08:29.855 --> 00:08:48.155
I mean, maybe I'm a bit biased because I'm speaking with so many cool companies doing crazy things Like, for example, my buddies running things over at Submix, where they're kind of building this collaboration tool for basically an online DAW, but to the point where you can still run your DAW locally and just run all your audio through a collaborative cloud tool.
00:08:48.155 --> 00:09:00.710
And it's interesting thinking about in general what the future of music production workflows will look like with all these different improvements in some of the more technical details, such as mixing and mastering, collaborating with one another, generating sounds and inspiration.
00:09:00.710 --> 00:09:04.645
For me, it's kind of overwhelming sometimes to think about how much is changing.
00:09:04.645 --> 00:09:13.851
I'm also a hardware person for the most part, aside from using some samples here and there, so I like to stick to my trusted local no screen type of thing.
00:09:14.039 --> 00:09:18.921
You want to be able to physically get a patch cable and stick it into that slot to make a different sound.
00:09:19.100 --> 00:09:19.441
Exactly.
00:09:19.441 --> 00:09:34.139
I mean that's for me, but still, it's interesting to see how many new music producers are coming to the game these days and all of the tools they have readily available some even free to already jumpstart their experience of producing music.
00:09:34.139 --> 00:09:35.159
I think it's super fascinating.
00:09:35.421 --> 00:09:38.769
Yeah, I think workflow is the main key here.
00:09:38.769 --> 00:09:44.414
With most of these tools, I think a lot of the debate is around creativity and are we replacing humans?
00:09:44.414 --> 00:09:56.575
And that kind of thing, which is valid, but I think really, when you see it as a tool to help you get from A to B or zero to 100, it's like what steps along the way can I take to make this quicker or better or easier, etc.
00:09:56.575 --> 00:09:58.908
So that's how I think it all fits in.
00:09:58.908 --> 00:10:07.251
Final question for me Do you specialize in any type of genre, or is this for a specific type of producer?
00:10:07.251 --> 00:10:08.193
How does that look for you?
00:10:08.259 --> 00:10:08.780
Good question.
00:10:09.341 --> 00:10:10.764
So long term vision.
00:10:10.764 --> 00:10:15.594
Our tools could be used by anybody, in any genre, in any location, from any level.
00:10:15.594 --> 00:10:20.010
But we can't start there because then we'll spend our whole lives developing and we'll never get anything done.
00:10:20.010 --> 00:10:23.562
So we are starting with, first off, a focus on electronic music.
00:10:23.562 --> 00:10:25.543
That's kind of our main market up front.
00:10:25.543 --> 00:10:35.109
We're also planning to only kind of target German and US users up front, then expanding to the UK and other EU regions, just because that's kind of our home lands.
00:10:35.109 --> 00:10:50.586
That advanced producers would still benefit from this, but they're the ones who more likely have a decent system set up already in terms of sample management and sample digging.
00:10:50.586 --> 00:10:52.113
For sure they could still save time.
00:10:52.113 --> 00:11:04.586
But I feel that veteran producers or beginner producers will benefit the most from the tool in terms of time saving, and they will also be the people who are in general more willing to modify their workflow and play with a tool like ours.
00:11:04.726 --> 00:11:05.006
Cool.
00:11:05.006 --> 00:11:06.951
Well, thanks for taking us through the tool.
00:11:06.951 --> 00:11:10.009
I think it's like my personal take is learning more about it.
00:11:10.009 --> 00:11:13.414
I was more and more impressed at the niche that you found.
00:11:13.414 --> 00:11:37.451
First off, the one shot thing, versus either generating a whole song or going to a library, and then the kind of tweakability aspect of it, which is OK start from a place and then, as you said, maybe you can generatively or iteratively expand on what you've done and just kind of get to that perfect sound which may end up being, like, you know, 0.1 second sample in your whole tune.
00:11:37.451 --> 00:11:41.515
Right that maybe it's repeated 100 times, but you know, I think that's that's really interesting.
00:11:41.515 --> 00:11:43.163
And then the workflow part of it.
00:11:43.163 --> 00:11:45.690
So I'm very excited for seeing how this comes together
00:11:46.051 --> 00:11:46.351
Me too.
00:11:46.371 --> 00:11:46.994
All right.
00:11:47.120 --> 00:11:52.413
So let's move on to some of the music that has inspired your tool and inspired the business.
00:11:52.413 --> 00:11:56.650
And the first track is not even a track, it's a whole damn album.
00:11:56.650 --> 00:12:04.433
Max, I think you're the first guest to go for an album as a track, which is kind of cool, and it's a good one.
00:12:04.433 --> 00:12:08.666
It's from the Tribe, it's People's Instinctive Travels and the Paths of Rhythm.
00:12:08.666 --> 00:12:12.322
This is from 1990, a seminal album from Tribe Called Quest.
00:12:12.322 --> 00:12:14.187
Tell us a bit about what this means to you.
00:12:14.408 --> 00:12:18.202
First off, sorry to have an aside but interesting that I'm the first person to pick an album.
00:12:18.202 --> 00:12:20.885
But I wanted to say in general, I really love this format.
00:12:20.885 --> 00:12:23.509
I think it's a super cool way to tell your story via songs.
00:12:23.509 --> 00:12:25.672
So big props to that idea.
00:12:26.933 --> 00:12:39.181
Yeah, this Tribe album for me and for our company I would say is super interesting because it's kind of the genesis for both my own personal music taste and also our company's direction in general.
00:12:39.181 --> 00:12:48.464
Growing up in California in the 90s after this album came out, I wasn't really exposed to this sort of music so much I would say growing up.
00:12:48.464 --> 00:12:49.787
I mean hip hop.
00:12:49.787 --> 00:12:54.490
There was some influence in the late 90s for me, but nothing on this level, I would say.
00:12:54.490 --> 00:13:00.049
And I still remember the time my friend first showed me this album Shout out Dennis Sippen and AJ Heslip if you're watching.
00:13:00.049 --> 00:13:00.701
Thank you very much.
00:13:00.701 --> 00:13:04.225
I remember listening to it the first time and thinking what is this Like?
00:13:04.225 --> 00:13:10.455
What is going on there from a production level and a lyrical level as well.
00:13:10.455 --> 00:13:20.205
It kind of just blew me away and in general it was also my first introduction to sample-based production, which again comes into Just 4 Noise as well.
00:13:20.205 --> 00:13:33.955
And for me this introduced me to a lot of the other big influences in my life in hip-hop, like Tupac, biggie Wu-Tang, Madl ib and Doom, the Alchemist, and other producers who in general use samples as a part of their production workflow.
00:13:33.955 --> 00:13:39.280
So for me that was just a big inspiration and introduction to a whole world of music that I wasn't really familiar with.
00:13:39.280 --> 00:13:42.961
I was also, at that time, more so playing with traditional instruments.
00:13:42.961 --> 00:13:50.586
I was playing piano, saxophone, trombone, trumpet, all these different instruments, and so learning this whole other world was super interesting for me.
00:13:50.995 --> 00:13:56.147
And Just 4 Noise, we kind of use this album as a reminder or a centering tool.
00:13:56.147 --> 00:14:01.562
I would say, like we use it to tell ourselves what are we doing this for, kind of like who are we building this for?
00:14:01.562 --> 00:14:09.336
And we like to imagine if we would have existed in the late 80s, would tribe have, uh, used a similar tool?
00:14:09.336 --> 00:14:11.986
Would they have benefited from a similar tool right now?
00:14:11.986 --> 00:14:25.299
Obviously, a lot of their sampling done in this album is, from you know, funk and soul records from like the 80s and 70s, but nonetheless, kick sampling and and using drums from other platforms or other creators was still a thing around that time.
00:14:25.299 --> 00:14:28.744
So we like to think if we existed back then, would we have been able to save them time?
00:14:28.865 --> 00:14:39.447
And it's also interesting looking at this album on the other side of things, because while it has a lot of samples based in it, it's maybe been sampled more than it has used samples, which is also super interesting, right.
00:14:39.447 --> 00:14:47.403
So also thinking about future workflows right For music producers who are trying to sample music.
00:14:47.403 --> 00:14:49.490
Would they be able to benefit from our tool rather than having to dig and grab copyrights to grab?
00:14:49.490 --> 00:14:51.938
You know Ali's kick drum used in Can I Kick it, or something you know.
00:14:51.938 --> 00:14:52.902
So it's yeah.
00:14:52.902 --> 00:14:58.080
For us it's just a super interesting album and also just a fun one to listen to.
00:15:22.480 --> 00:15:26.849
Yeah, I mean there's a lot to unpack from this album that go in lots of different directions.
00:15:27.056 --> 00:16:05.104
When you were talking about aiming Just 4 Noise at bedroom producers at the beginning, I immediately thought of Q-Tip and how a lot of the genesis of the album came from him making pause tapes, which I guess in a way you could say that was one of the first consumer level sampling tools, which is just a cassette deck, and I think he probably spent hours and hours pressing pause after every bar of looping something and then going back and doing that and thinking how incredibly manual that process was, but using the tools available and how we've got to a point and I'm sure I'm not sure what happened post-production.
00:16:05.495 --> 00:16:11.884
They then later resampled stuff into an MPC or something and looped it in a more, I guess, digital way at the time.
00:16:11.884 --> 00:16:22.458
But the hours and hours that he would have spent not just doing the actual looping but practicing, I'm sure if you mess up you have to rewind the tape, go back to the pause point, re-cue the record or the tape or whatever, and just do it again and again.
00:16:22.458 --> 00:16:24.421
Recue the record or the tape or whatever and just do it again and again.
00:16:24.421 --> 00:16:40.245
So it's incredible the access, I suppose, to the ability to sample, which at that time would have been open to very, very few people, either professional studios or I guess some people might have had an MPC-60 or an SP-1200 at the time.
00:16:40.245 --> 00:16:54.261
But yeah, just the idea of going from these pause tapes that created one of the most influential hip hop albums of all time to the fact that someone could today say like, hey, give me a kick drum that sounds like X, right, and then, in seconds, I guess, get something out of the other end.
00:16:54.402 --> 00:16:55.325
Yeah, it's super interesting.
00:16:55.365 --> 00:17:06.585
I mean, when you think about the craft of sampling in general and how it's evolved, I think there's a lot of people around these days still producing music who would see what we're doing and say you're ruining the craft.
00:17:06.585 --> 00:17:20.125
Back in the day I was ripping at tapes and I was going through all this manual effort like crazy lows and crazy frustration, but also crazy amount of gratification when I get it right to the point where it's instant with what we're doing.
00:17:20.125 --> 00:17:22.458
That's kind of the whole selling point of what we're doing.
00:17:22.458 --> 00:17:24.775
So it's interesting to think about it from that perspective.
00:17:24.775 --> 00:17:33.242
But at the same time you always wonder if those tools were always available, would people even be complaining about the craft changing, right?
00:17:33.483 --> 00:17:40.038
I still think that there's a happy middle ground somewhere, right, in music production in general and, like I said, we're not trying to take away from the creative process.
00:17:40.038 --> 00:17:46.797
We don't want to make it so that you prompt every single aspect of your song, even though technically it's possible, right.
00:17:46.797 --> 00:18:00.934
But I think that there is a balance to be found in terms of using your creativity and your energy to build something that's important to you while also automating the parts that to you are less important, which for us is sample digging and sample usage.
00:18:01.276 --> 00:18:06.508
It's interesting as well the balance between what I would call the craft and the end product.
00:18:06.508 --> 00:18:14.828
So in a way, if you're making music for people, for an audience, and then they like the end result, does it matter?
00:18:14.828 --> 00:18:19.586
If you spend hours and hours creating, I know for certain fans that can be like a plus.
00:18:19.586 --> 00:18:21.762
Okay, I like the lore behind this album.
00:18:21.762 --> 00:18:27.499
I love the fact that you know Q-Tip spent hours digging in crates or raiding his dad's record collection and doing this.
00:18:27.499 --> 00:18:35.263
I think it does add to the story around an album but at the same time it doesn't necessarily make it better or worse to have done that.
00:18:35.404 --> 00:18:37.136
I think the craft is for the producer.
00:18:37.136 --> 00:18:41.395
So if you're a producer but just really likes doing a certain thing, cool, do it however you want.
00:18:41.395 --> 00:18:42.939
You can go hardware, only you can.
00:18:42.939 --> 00:18:43.980
You know full software.
00:18:43.980 --> 00:18:45.683
You can go ai this, that, any other.
00:18:45.683 --> 00:18:56.894
But also if you're making a product for someone I'm not sure how many people knew how that record was produced or was was thinking so hard about okay, what's the, the methodology behind it?
00:18:56.894 --> 00:19:00.343
They just heard it and it was instantly resonated with them.
00:19:00.343 --> 00:19:05.299
Right, that kind of craft versus product piece I think is an interesting one to think through.
00:19:05.441 --> 00:19:07.565
Yeah, I mean in the entire art world.
00:19:07.565 --> 00:19:12.101
Like you could talk about this all day, like what do real creators do?
00:19:12.101 --> 00:19:14.227
How is art perceived, valuable and all that stuff.
00:19:14.227 --> 00:19:21.616
Right, it's super interesting, so I won't begin diving too deep into that, but yeah, I think you hit it on the nail on the head.
00:19:21.616 --> 00:19:31.334
Like each artist needs to find their own craft and find the methods that they need to use in order to create something that, to them, is expressing something or another.
00:19:31.334 --> 00:19:33.880
Right, it's also up to them to decide what they want to communicate.
00:19:33.880 --> 00:19:46.541
But to the people who just hate on something because it's not how they do it, I mean, that's on them, in my opinion, to put it kindly yeah, there's a yeah, certain type of person that.
00:19:46.623 --> 00:19:48.006
So it's a sidebar in a way.
00:19:48.006 --> 00:20:01.406
But you know, like in the audiophile community, where people care more about the technical specs of like the speaker wire they use and the impedance, it's like, okay, you care about the equipment, but are you really listening to the music?
00:20:01.406 --> 00:20:04.663
It's that kind of mentality which I don't like so much.
00:20:04.663 --> 00:20:14.287
But yeah, I think the other bit you mentioned mentioned, which is interesting, that I'd like to go into a bit more, is this idea of resampling.
00:20:14.287 --> 00:20:25.020
So immediately, you know, you listen to Bonita Applebaum and then you hear the Fugees Killing Me Softly and there's a bunch of stuff that's been sampled off this album and subsequent Tribe albums and I always love this idea of that recontextualization.
00:20:25.161 --> 00:20:32.378
For a start, it's taking a bunch of samples from disparate genres different times, put them on a track and then someone else listened to that track.
00:20:32.378 --> 00:20:34.125
So I'm going to sample this section.
00:20:34.125 --> 00:20:36.580
So you're almost like layering and layering you mentioned.
00:20:36.580 --> 00:20:39.496
It kind of connects a bit to the idea of Just 4 Noise as well.
00:20:39.496 --> 00:20:44.082
What's your take on that and how do you think that continues to evolve into the future?
00:20:44.182 --> 00:20:55.819
Yeah, it's a super good point, um, and it's also like kind of as another side point, like, as you were talking about that, I was realizing that we're talking about hip-hop specifically, right, and how hip-hop kind of created this idea of using sampling as a tool.
00:20:55.819 --> 00:21:00.938
And now, when you look at the music industry in general, there's not a genre that doesn't touch it for the most part, right.
00:21:00.938 --> 00:21:08.843
I mean, maybe there's some super niche groups that have this like analog approach to production, of course, but in general it's kind of everywhere, right.
00:21:08.843 --> 00:21:22.880
And this idea of resampling to me is super fascinating because, as music producers create more and more content and in general, as more audio data becomes available in the world, you have this concept of I have the power in my computer to, for a very low price, create any sound imaginable.
00:21:22.880 --> 00:21:26.080
I have the power in my computer to, for a very low price, create any sound imaginable, right.
00:21:26.080 --> 00:21:41.425
But we still draw ourselves towards these small groups of sounds, whether that's a memorable chorus from a super popular song or a super memorable drum loop like a breakbeat, right, which has been maybe one of the most resampled drum sounds in the world, right.
00:21:41.425 --> 00:21:45.487
So it's really fascinating to me and I'm trying to dive deeper into the psychology behind.
00:21:45.547 --> 00:21:56.351
Why do we find ourselves always migrating or navigating towards these same sounds, rather than exploring this whole undiscovered world of new sounds that haven't been created or have been created but only used once?
00:21:56.351 --> 00:22:00.231
I think in general that's kind of what production workflows look like.
00:22:00.231 --> 00:22:05.574
When I want to describe a kick drum and I want to generate a kick drum, I'm not using completely new words to describe it.
00:22:05.574 --> 00:22:12.343
I'm using the same words that you would maybe use or somebody else would also use.
00:22:12.343 --> 00:22:13.708
So we're kind of instilling that mindset in our product.
00:22:13.708 --> 00:22:22.701
To the fact that, yeah, it's kind of hard to interpret natural language when describing a sound, because how I call a clicky 808 is maybe a little different how you would perceive a clicky 808.
00:22:22.701 --> 00:22:34.541
But still, we're trying to get as close as we can to understanding this language and getting to the point where we can always center around this one universal pool of sounds that music producers tend to navigate to.
00:22:34.541 --> 00:22:38.503
But it's not an easy task, but it's a fascinating journey, that's for sure.
00:22:38.695 --> 00:22:49.240
It is, but I'm sure there's some neuro-linguistic and neuro-scientific elements to how this all works, as well as the fact that you know, like some things are probably just nostalgic.
00:22:49.240 --> 00:22:56.673
You know like the amen break or something like people associate with different things, maybe like drum and bass and jungle or hip-hop.
00:22:56.673 --> 00:23:04.066
Maybe, like you say, you could create a future classic sample through synthesis and that kind of thing.
00:23:04.066 --> 00:23:05.166
But yeah, I have no idea.
00:23:05.326 --> 00:23:08.601
Basically, how it all works and why we like this stuff.
00:23:08.601 --> 00:23:09.042
Okay, cool.
00:23:09.042 --> 00:23:22.223
So I think we can move on to the second track, and you actually mentioned the idea of something becoming more popular and coming into the mainstream, and this is a pretty mainstream track for number two.
00:23:22.223 --> 00:23:24.804
So this is Toxic by Britney Spears.
00:23:25.356 --> 00:23:27.576
Let us know about why this is significant to you
00:23:27.738 --> 00:23:32.036
it's so funny, um, going from Tribe to Britney, and I did it on purpose, just because it's funny.
00:23:32.036 --> 00:23:33.579
But I mean first things first.
00:23:33.579 --> 00:23:38.980
Again, I said I was growing up in the 90s and the song was a banger, obviously as a, as a, as a child, I mean.
00:23:38.980 --> 00:23:42.976
It was a hugely popular song and music video as well.
00:23:42.976 --> 00:23:48.941
But but the story behind it is actually has little to do with the song and more has to do with our company, which is really funny.
00:23:48.941 --> 00:24:02.847
So my co-founder and I, Henning, along with our full stack engineer, who's now also on board, were previously working at a company together and we had a hackathon where we had 24 hours to build out an idea for AI and creativity.
00:24:02.847 --> 00:24:08.548
That was kind of the concept and we built out this crazy ridiculous as a joke.
00:24:08.548 --> 00:24:18.948
Of course, AI takes over humans creative platform where you don't have to do anything and AI does song generation, remix, creation, DJing and visuals for you.
00:24:18.948 --> 00:24:39.174
We obviously, within 24 hours, built barely any of it, but our biggest selling point was that you could prompt the remix of a track via style transfer, which has now become a super popular synthesis or a neural synthesis mode that a lot of people are looking into, especially when it comes to timbre transfer.
00:24:39.174 --> 00:24:44.265
When you say, here's a piano, make it sound like a saxophone, right, which is super cool.
00:24:44.365 --> 00:24:55.788
But we, we took this concept in the remixing world so you can feed a song into this engine and say make this in this genre or this style, and then it would interpret your text and then create a new song.
00:24:55.990 --> 00:25:12.934
We took Britney Spears' Toxic and made it into like a 90s industrial, hard style techno track and, honestly, we ended up winning this hackathon and I think this song is the reason why we did, because it just sounds so good, like such a genuinely great.
00:25:12.934 --> 00:25:19.147
I'll send it to you after um and maybe like maybe there's no copyright there so you could play it.
00:25:19.147 --> 00:25:19.817
No, probably not.
00:25:19.817 --> 00:25:31.499
But it's funny because this was probably a year and a half ago or two years ago when we did this hackathon, and this was just us playing around thinking that there's cool things you can do in the, in the art world and music world specifically.
00:25:31.499 --> 00:25:42.077
So it was funny how, a year and a half later, we find ourselves working full-time on something that's obviously very different, but within the same world of neural synthesis, as we call it, and music, ai, whatever you want to call it.
00:25:42.077 --> 00:25:45.281
So that's kind of of the story of Toxic by Britney Spears.
00:25:46.242 --> 00:25:47.064
That's fun.
00:25:47.064 --> 00:25:49.066
Why did you pick Toxic of all tracks?
00:25:49.205 --> 00:25:56.384
So once we were done with all the building, we were probably at that point 13 or 14 hours into the hackathon and it's like 4 am.
00:25:56.384 --> 00:26:04.063
We're like at this point, we're like having drinks, you know, like relaxing a little bit, and my co-founder, henning, was like what song should we do?
00:26:04.063 --> 00:26:08.980
And I was like toxic, easy, for some reason it was just already on the top of my head.
00:26:08.980 --> 00:26:21.060
We did a few other ones like old 90s rave stuff in like the style of like american bluegrass, but it didn't hit as hard as as um industrial techno toxic.
00:26:21.201 --> 00:26:22.465
It was a funny experience nonetheless
00:26:22.847 --> 00:26:38.858
it's an interesting choice because you know, earlier we were talking about the method, method versus end product thing and I remember, like you, Toxic coming out as a track at the time and I didn't know there was a Bollywood sample on it as like the main hook.
00:26:38.858 --> 00:26:53.355
And this is like when Timbaland was doing more of that kind of sampling of kind of Indian and Middle Eastern stuff, so it was coming en vogue but it was really disguised as like a pop song that had a sample and that was like interesting.
00:26:53.355 --> 00:27:04.182
But you, you picked that because it is almost shows that synthesis or the incorporation of sampling into popular music in a way that I don't know.
00:27:04.182 --> 00:27:07.191
Do most people know there's a sample on there?
00:27:07.191 --> 00:27:08.895
I'm not sure probably not.
00:27:08.977 --> 00:27:21.039
I mean, it's not the the most famous um early 2000s, like middle east or Bollywood sample used in a pop track, but still, I think it's kind of a mealable thing showing where it came from and showing the music video.
00:27:21.039 --> 00:27:38.092
But it is super true and it loops back to what we were saying before, right, I mean it started off in hip hop and more traditional analog style music production and then already within 10 years or 15 years it was everywhere, right and super accessible at that point as well, which is also super interesting
00:27:38.653 --> 00:27:49.878
okay, so producer challenge concept for you could you recreate britney spears is toxic, using Just 4 Noise, or, by the way?
00:27:49.919 --> 00:27:52.932
if you were to try and do that, how would you start going about that?
00:27:53.132 --> 00:27:53.833
that's really funny.
00:27:53.833 --> 00:27:57.660
Um, I mean the vocals, no, no shot.
00:27:57.660 --> 00:28:02.292
I mean maybe eventually, but I doubt we're going to get into that.
00:28:02.292 --> 00:28:16.196
I mean again, our, our um mo is one-shot samples, right, and, and you can take a um, you can generate a one-shot sample of a synth, right, and then plug it into a sampler and play chords and melodies and all that stuff.
00:28:16.196 --> 00:28:19.164
So I would say, for the most part, anything is achievable.
00:28:19.164 --> 00:28:25.501
It's definitely taking much more time at that point right, as opposed to generating long like short-term loops or stuff like that.
00:28:25.501 --> 00:28:32.694
But right now, where we are in terms of the development of the product itself is first we're focusing on drum sounds, so we already got the kick drum down.
00:28:32.694 --> 00:28:35.700
Then we're moving over to snares, hi-hats, toms and so on.
00:28:35.700 --> 00:28:45.672
Then we plan to move to probably more classical synthesizers and traditional instruments, for example strings, a 303, you know that and that's sort of the direction.
00:28:45.672 --> 00:28:50.250
Then we plan to go deeper into what other directions our users want us to go.
00:28:50.250 --> 00:28:54.303
But that's definitely for, for our side, a while down the road got it.
00:28:54.343 --> 00:29:13.414
so I guess the step one would be okay, I can create the individual drum hits that would probably approximate that type of sound and then perhaps next you could generate like the bass, like let's give me like a middle c in this, you know particular, like 303 style or whatever, and you could play that across the keyboard.
00:29:13.414 --> 00:29:25.193
And then I future, future thing, if you ever go to that space, would be then, okay, create me a you know four second Bollywood style string sample or something, and then you could kind of yeah, okay.
00:29:25.654 --> 00:29:26.957
Exactly, exactly.
00:29:26.957 --> 00:29:40.521
But it's so interesting to think about using our tool for the entirety of the of the song creation process because obviously it's it's possible, but I don't really imagine people using it for for that way.
00:29:40.521 --> 00:29:46.715
As I said, I feel like we're much more of a utility than like an end to end creative platform or something Right.
00:29:46.715 --> 00:29:59.336
I mean, it could be interesting to see what that might look like one day, and obviously a lot of other companies are trying to achieve a similar thing in terms of entire song generation or creating entire stems for tracks, but I don't really see us going in that direction.
00:29:59.336 --> 00:30:06.890
I see as much more as a side tool that a creative person can use to cut down their time in order to focus on their creative aspects of their process 100.
00:30:07.371 --> 00:30:09.416
I think this comes back to the workflow thing.
00:30:09.416 --> 00:30:14.340
It's probably pretty inefficient to try and create a whole entire track from these one shots.
00:30:14.340 --> 00:30:21.000
You probably like 80 of a track there and you're like, okay, I want to tweak this type of sound or, you know, get something in.
00:30:21.000 --> 00:30:22.202
So, yeah, that makes a lot of sense.
00:30:22.202 --> 00:30:27.402
Okay, so we're gonna move on to the final track and actually here is another first.
00:30:27.402 --> 00:30:37.440
So it's not an album, but it's basically I want to call it a set and I'm gonna leave you to describe this in more detail because a lot going on in this particular video.
00:30:37.440 --> 00:30:40.654
So, just for everyone listening, max sent me a video.
00:30:40.654 --> 00:30:41.537
He's in it.
00:30:41.537 --> 00:30:49.079
It's basically a tiled DJ booth type situation and I think DJ set doesn't quite do it justice.
00:30:49.079 --> 00:30:51.971
So, yeah, I'll hand it over to you, Max, to describe this in a bit more detail.
00:30:52.011 --> 00:30:56.077
Sure, sure, as I told you, it's kind of like a little selfish promo, but also there's a story behind it.
00:30:56.077 --> 00:30:57.721
This is me and my partner.
00:30:57.721 --> 00:30:58.501
We have a DJ name.
00:30:58.501 --> 00:30:59.564
It's called Kotik Merkotik.
00:30:59.564 --> 00:31:00.924
It's a Ukrainian name.
00:31:00.924 --> 00:31:03.898
We played this radio show which is kind of popular in Berlin.
00:31:03.898 --> 00:31:07.029
It's called HÖR, which means listen in German.
00:31:07.210 --> 00:31:14.700
They record it in like this tiled bathroom looking thing and everybody jokes that it's a bathroom, and this particular set was what I would call a hybrid set.
00:31:14.700 --> 00:31:23.323
So half of it is me playing around with this modular synthesizer and the little drum machine and the other half is my partner playing tracks off CDJs.
00:31:23.323 --> 00:31:35.310
The reason why I chose this video is because of the drum machine in there and there's a bit of lore behind this drum machine and Just 4 Noise, that drum machine I actually got at the same time I did the hackathon with my company.
00:31:35.310 --> 00:31:36.811
I had the company buy it for me.
00:31:36.811 --> 00:31:37.172
Actually.
00:31:37.172 --> 00:31:39.354
Thank you very much for the drum machine.
00:31:39.354 --> 00:31:47.759
It's super basic Korg Volca Beats, but very reliable drum machine and I was essentially relying on that thing in my production workflow.
00:31:47.759 --> 00:31:58.972
I started only using that and stopped digging for samples, because I was always using drum samples in my workflow and I hated it.
00:31:58.972 --> 00:32:08.849
I hated always going through Splice and buying sample packs and not using 95% of them and then losing them and having to dig and find where my folders are and all that stuff.
00:32:08.849 --> 00:32:15.077
To be fair, I probably wasn't the most organized person in my folders, so maybe that's on me, but I just hated the process.
00:32:15.077 --> 00:32:18.532
I wanted to just play around with knobs and forget about everything.
00:32:18.955 --> 00:32:31.660
I was going through a phase where every single morning I would wake up early and I would jam and this was kind of my winter Berlin routine to ignore the gray skies and kind of start the day off on a positive light, although sometimes it was frustrating.
00:32:31.660 --> 00:32:37.156
But there was one day where I turn on all my gear and I go to turn on my drum machine and it just broke for some reason.
00:32:37.156 --> 00:32:37.959
I don't know.
00:32:37.959 --> 00:32:38.720
I still don't know why.
00:32:38.720 --> 00:32:42.285
Actually I haven't gotten it repaired still and it broke.
00:32:42.730 --> 00:32:52.827
And that was kind of the day that I realized Just 4 Noise oise needs to exist, because I spent the rest of my 45 minutes that I usually set aside for this just trying to find my samples.
00:32:52.827 --> 00:33:00.482
I couldn't find my folders again and I had already deleted my Splice account at that point and I didn't end up creating anything because I didn't want to use presets.
00:33:00.482 --> 00:33:07.118
I had already had this idea of Just 4 Noise before and I'd already talked with my current co-founder about it a few times and we liked the idea.
00:33:07.118 --> 00:33:12.276
We knew there was a pain point out there, but after this experience for me it was like OK, that's it, let's do it.
00:33:12.276 --> 00:33:14.480
If I had this right now, I wouldn't have to worry.
00:33:14.480 --> 00:33:19.910
I could just throw a high quality drum sample there and go on playing with my knobs.
00:33:19.910 --> 00:33:27.963
So it's just funny how that little drum machine, I would say, kind of kickstarted us to finally just get our hands dirty and start doing it.
00:33:28.471 --> 00:33:30.317
I love the severity of a situation.
00:33:30.317 --> 00:33:33.299
It's like my drum machine broke so I founded a whole company.
00:33:33.299 --> 00:33:36.760
I made a whole tool to make my drum samples from.
00:33:38.289 --> 00:33:43.700
Yeah, I mean, realistically, we would have probably done it eventually, right, we were talking about it a lot at that point, but more so like oh, wouldn't this be cool?
00:33:43.700 --> 00:33:48.257
And oh, we could actually do that, and oh, let's read this research paper and see if it's possible.
00:33:48.257 --> 00:33:50.451
But this, this specific day, was a day I called him.
00:33:50.451 --> 00:33:52.493
I was like, dude, meet me at the cafe.
00:33:52.493 --> 00:33:53.856
Like we need to talk about this now.
00:33:53.856 --> 00:33:56.981
Like let's start doing it, you know this, is it?
00:33:57.522 --> 00:34:00.055
so there's a few things to to touch on in that.
00:34:00.055 --> 00:34:00.676
And you?
00:34:00.676 --> 00:34:08.510
I'm trying to go back to something you mentioned at the beginning of the session, which was about file management, this kind of concept of keeping stuff organized.
00:34:08.510 --> 00:34:20.391
So me personally, I hate that idea because usually when you're trying to grab a file or grab a sample, you're in the middle of a creative flow, so you're not going to think, oh, I'm going to rename this, this, I'm going to put it in this folder.
00:34:20.391 --> 00:34:24.440
So how does Just 4 Noise start to address that issue?
00:34:24.560 --> 00:34:29.641
It's a really good question and I'm sorry but I'm not going to spill all the magic here, because we have some really nice ideas.
00:34:29.641 --> 00:34:36.981
Sample management and MIDI management and overall data management for music producers should be its own product.
00:34:36.981 --> 00:34:45.036
I mean, I know some cool people building cool things like AudioCypher shout out Ezra building super cool MIDI management stuff and I think it's a whole other world.
00:34:45.036 --> 00:34:50.018
But we also have some really nice ideas of how we can kind of improve that, at least for one-shot sample management.
00:34:50.199 --> 00:35:03.902
Aside from that, in general, I feel that by having a synthesizer right, a neural synthesizer that can generate one-shot samples for you, you semi-remove the need to have structure right.
00:35:03.902 --> 00:35:11.398
I mean, to be fair, you're going to find some samples by interfacing with Just 4 Noise, where you're like I want to use this every single time and I need to save this.
00:35:11.398 --> 00:35:26.121
But I have a theory that if you have a neural synthesizer on your side, or even a drum machine on your side, you don't need as much clutter, you don't need as much data in order to arrive at something that you think is good enough for your projects and good enough for your, your own personal sound.
00:35:26.121 --> 00:35:35.675
Even aside from file management, by using neural synthesizers for one-shot samples, you're able to reduce the amount of samples you need throughout your creative process.
00:35:35.996 --> 00:35:45.474
That's kind of my mindset, at least so I I suppose that's kind of the synthesis mindset, which is you've got the tool, so to speak, to create this, the stuff.
00:35:45.474 --> 00:35:56.815
You don't necessarily save the samples from it, or obviously you can track it out and stuff, but each time you're coming to it and say, okay, I want this type of sound and I'm gonna make these things happen to create that thing.
00:35:56.815 --> 00:36:02.083
So it's not like I need 10 versions of that, I just will make it live basically each time exactly.
00:36:02.704 --> 00:36:12.016
The other little thing and I'll mention this without going too deep into it is, as you generate a sound with our tool, you're already labeling that sound right.
00:36:12.016 --> 00:36:23.152
When I say I want this type of kick drum with these characteristics, it's saved with those characteristics right, so just planting that on you.
00:36:23.152 --> 00:36:25.378
You can think about the different ways in which digging becomes easier when you have tagged data.
00:36:25.398 --> 00:36:29.253
Yeah, like searching not just by the type but by the different characteristics.
00:36:29.253 --> 00:36:35.574
I suppose, when it comes to the neural synthesis part, can it create the same thing twice.
00:36:35.574 --> 00:36:41.994
If I said, oh, I really liked when I made this kick drum, but I want to make it again, would the same prompt make the same thing, or is it going to be like slightly different?
00:36:42.054 --> 00:36:49.092
It's a really funny question and like it's interesting because so many people ask us this question and I never really thought about it until people started asking it so much.
00:36:49.092 --> 00:36:55.438
In theory, the way the model is built, you can put this kind of hot and cold range on it.
00:36:55.438 --> 00:37:06.992
I don't know if you've seen in, like some of the chatbots out there, you can decide how wild the responses are based off this hot and cold meter and you can essentially control the machine learning models that we're building in the same way.
00:37:06.992 --> 00:37:12.244
So you can set your model to hot, to where every single time you prompt the same exact search.
00:37:12.244 --> 00:37:15.659
You're going to get something completely different in this realm based off those terms.
00:37:15.659 --> 00:37:23.014
Or you can set it very cold, to the point where you get nearly the exact thing every single time.
00:37:23.034 --> 00:37:27.871
But in theory, how this neural process works, it's never going to repeat the same exact action every single time.
00:37:27.871 --> 00:37:32.490
So with the same words, technically the end result will also be different.
00:37:32.490 --> 00:37:39.259
Just as if you use an image generator or a video generator and you give it the same prompt, it'll always at least be slightly different.
00:37:39.259 --> 00:37:55.726
Right, it'll maybe still like like a Picasso painting or whatever, because you said Picasso, but some characteristics are always a little different, and that's the fascinating part about stable diffusion in general, which is one of the pieces and and almost all image and audio generated stuff audio generated stuff.
00:37:56.289 --> 00:38:06.115
You know, I guess the interesting thing about that is it's almost like coming back to analog, where everything is like slightly different, like if a temperature is a bit different in the room, it's not stable necessarily.
00:38:06.115 --> 00:38:11.635
Yeah, that's a little mini revelation I've just had, because the files obviously stay the same every time.
00:38:11.635 --> 00:38:17.599
But if you're doing slightly different things on the wildcard front, you're going to come up with slightly different results.
00:38:18.141 --> 00:38:18.831
Exactly so.
00:38:18.831 --> 00:38:19.976
We definitely want to find the balance.
00:38:19.976 --> 00:38:32.757
But when we talk about file management, we want to find the balance between persisting nice sounds that you generate and you want to keep, while also giving you an option to create iterations of that sound via further prompting.
00:38:32.757 --> 00:38:33.862
That's kind of our idea.
00:38:34.244 --> 00:38:34.445
Cool.
00:38:34.445 --> 00:38:39.362
Well, Max, it's been great to talk through your stories and your business with you.
00:38:39.362 --> 00:38:43.474
I've learned a few things and I'm probably going to go away and research a bit more after this.
00:38:43.474 --> 00:38:48.193
But before we wrap up, where can people go to find out more about Just 4 Noise?
00:38:48.454 --> 00:38:50.259
Yeah, first off, super fun, Tom.
00:38:50.259 --> 00:38:51.402
Again, I would love the format.
00:38:51.402 --> 00:38:53.913
Really enjoyed being on the program, so thanks for having me.
00:38:53.913 --> 00:39:00.742
We will be having an official launch early next year, probably around late winter, early spring, and you can sign up on www.
00:39:00.742 --> 00:39:00.742
Just 4 Noise.
00:39:00.742 --> 00:39:09.298
com with the number four no periods, no spaces or anything like that or www.
00:39:09.298 --> 00:39:09.298
just-noise.
00:39:09.298 --> 00:39:15.635
com Tom will shoot the link somewhere for sure as well.
00:39:15.635 --> 00:39:25.679
And also, as another little side plug, I will be releasing an album early next year, so you can also follow me and Kotik Merkotik, who's my partner, to see what music we're putting out.
00:39:26.130 --> 00:39:29.315
Awesome, max, been a pleasure Great speaking with you.
00:39:29.570 --> 00:39:30.956
Right back to you, Tom, best of luck.
00:39:38.342 --> 00:39:38.681
Ciao