[00:00:00] Diaa El All: We've done it. Like, there is a way where technological advancement and working with the rights holders and protecting them can co-exist together.
[00:00:10] Dan Blumberg: That's Diaa El All, and he's the CEO and founder of Soundful, which enables anyone to create brand new music in just a few seconds with AI.
[00:00:19] Diaa El All: People that don't know how to produce. And they're great singers. They're great songwriters that don't know how to put a fully composed track together. We help them do that.
[00:00:30] Dan Blumberg: And Soundful is making music in a responsible way. Soundful doesn't just train its models on whatever music is on the Internet. Instead, they created proprietary data by painstakingly recording musicians playing one note at a time.
[00:00:42] Diaa El All: A kick drum is sampled once, a C note on a piano is sampled once. From there, we train the model and it will take these individual sounds and layering them together the same way that a musician would do it.
[00:00:54] Dan Blumberg: On this episode, we're literally making music, and we're exploring a fascinating marriage of [00:01:00] human and machine.
[00:01:00] Dan Blumberg: Welcome to CRAFTED.., a show about great
[00:01:11] Dan Blumberg: products and the people who make them. I'm Dan Blumberg. I'm a product and growth leader. And on CRAFTED., I'm here to bring you stories of founders, makers, and innovators that reveal how they build game changing products and you can too.
[00:01:29] Dan Blumberg: CRAFTED. Is produced in partnership with Docker, which helps developers build, share, run, and verify applications anywhere without environment confirmation or management.
[00:01:39] Dan Blumberg: Docker's suite of development tools, services, and automations accelerate the delivery of secure applications. Learn more at docker. com. And CRAFTED. Is produced by Modern Product Minds, where I advise companies on product, discovery, growth, and experimentation. Learn more and sign up for the CRAFTED. Newsletter at modernproductminds.
[00:01:58] Dan Blumberg: com. [00:02:00]
[00:02:01] Dan Blumberg: Diaa was just named to Billboard's 40 under 40 list, and it's in part thanks to the way Soundful is building its AI ethically and with the support of the music industry. I started our conversation by asking him about a recent letter that was signed by top artists such as Billie Eilish, Nicki Minaj, Stevie Wonder, R.E.M., and the estates of Bob Marley and Frank Sinatra. The letter states, this assault on human creativity must be stopped. We must protect against the predatory use of AI to steal professional artists voices and likenesses, violate creators rights, and destroy the music ecosystems.
[00:02:36] Diaa El All: I'm a hundred percent standing behind that.
[00:02:39] Diaa El All: I really think that we should be very careful about like the advancement of technology. overstepping over the rights holders, whether that's, you know, the artists or the producers or any creative works, we should be asking for licensing, we should be working with them to develop the business model and the technology moving forward.[00:03:00]
[00:03:00] Diaa El All: And it is very saddening, actually, that, There is a lot of the companies out there that are taking the easier and the short route of going out and, you know, scrubbing the internet and training on copyrighted materials. And they are in the, in the thought of like, don't ask for permission, ask for forgiveness later.
[00:03:18] Diaa El All: That's really wrong. And it's setting the precedent for, you know, the next generations to just be continuing and following in this footsteps. I really believe like now more than ever that the labels and the artists and the creative community are willing to work together with the, with AI companies to actually develop likeness, models, et cetera.
[00:03:42] Diaa El All: It has to be done with them, not just taking the easier route. And later on, we'll figure it out. I really truly believe it's kind of like. the Napster moment and Spotify. Napster just gave everything out there and Spotify tried to figure out a business model, but [00:04:00] actually it is worse than the whole Napster part.
[00:04:02] Diaa El All: Why? Because it's creating derivative of creative work that I have put a lot of work into it. And it's just creating a thousands or hundreds of thousands or millions of different permutations of it using my sounds, using everything that I've done. So it is a, an inflection point in the industry right now that The tech world and the, and the music worlds are going against each other.
[00:04:24] Diaa El All: Um, and I'm sure you've seen like the TikTok, like one of the biggest argument about TikTok. It's not just like licensing. It's about using the catalogs that TikTok has and training their own AI models without the permission. And I truly, truly believe we've done it. Like there is a way where, you know, technological advancement and working with the rights holders and protecting them can coexist together.
[00:04:49] Diaa El All: And, you know, we've done the deals with these labels and the artists that proves that they're actually open for that.
[00:04:57] Dan Blumberg: [00:05:00] Diaa has been a musician since his parents got him a piano at the age of three. He went from Cairo to the Royal Academy of London to study classical piano, then moved to the U. S. to make a career out of music. He worked as a sound designer, a touring artist, a DJ, and a producer.
[00:05:17] Diaa El All: My first gig, I remember getting paid, $50.
[00:05:21] Diaa El All: And as a kid I did the math of like, oh my God, I can be playing every night and if I can make, just bring it up to a hundred dollars or $150, I'll be making $3,000 a month. Yeah, that that would be amazing.
[00:05:34] Dan Blumberg: Yeah.
[00:05:35] Diaa El All: But came to learn that there is. So much competition and I had to wear a lot of different hats.
[00:05:40] Diaa El All: I was a promoter. I was passing flyers out in the malls. I was signed to record labels here in the U. S. I was signed to record labels out in, in Europe, in the Middle East. And my career shifted into ghost production. And this is when I went the entrepreneurial route, uh, started [00:06:00] multiple ventures, successful exits along the way.
[00:06:03] Diaa El All: And the one that resulted into Soundful was it's a company called Smart Boost, which is a digital marketing agency that we started three guys in the garage, but we wanted an edge into the space. And we started developing our data science team, software engineering team to develop tools to Automate a lot of the campaigns that we were running and that gear feels the edge into the space to work with some of the biggest brands in the world from GoPro to Ferrari to multi billion dollar SAS, B2B companies.
[00:06:36] Diaa El All: And that was really being exposed to what the power of the machine is. This is when Soundful came about.
[00:06:45] Dan Blumberg: I love that. And that makes a ton of sense that that's where you saw the power of AI, because that's where that's really where good old fashioned AI had a huge mark, right? Was in marketing automation.
[00:06:54] Dan Blumberg: And so how did, how did you then get from that good old fashioned AI and marketing work [00:07:00] to, Oh, wait, we can make music with this.
[00:07:02] Diaa El All: One of my co founders, Josh had a website that you can go on and you can hire ghost producers and they can create tracks for you, but he had major scale and he was wanting to build different tools that automate the way of a producer actually creating loops or creating sections of tracks.
[00:07:20] Diaa El All: So developed an, an algorithm that can create loops or just sections of tracks. And given my background, he knew me know what was, um, what my involvement is into that side of the space. And he brought it over to me and. I saw a little bit something different that not just like a tool for musicians or, you know, producers, this is a tool for that can be leveraged by content creators, by brands, for TV and film space.
[00:07:50] Diaa El All: The way that we would be able to achieve that is to maintain the highest quality possible, and that we can create an output that can trick the human ear to believe it was done by a human. [00:08:00] And this is when the idea of Soundful was born.
[00:08:03] Dan Blumberg: Cool. Well, I'd love to see, I'd love to play with it. Do you want to call it up?
[00:08:07] Diaa El All: Absolutely.
[00:08:08] Diaa El All: So when we were developing Soundful, we were after democratizing music creation. So when we were taking consideration building the UI, we wanted to simplify it so anybody can go into the platform and with a few clicks, they're able to create a studio quality output. So on the left hand side with Soundful, we have the navigation bar, Genres, collections, uh, creating loops.
[00:08:33] Diaa El All: And we also have a global tracks library that is auto generated by our platform. On the homepage, there's three actions. You sample what's behind it, you favor it, or there's this red button. Anywhere you see the red button on the platform, it means creation. So if I want to create something from scratch, click create a track.
[00:08:51] Diaa El All: You're selecting the overarching genre. Let's say EDM, and then you'll be driven to different [00:09:00] styles of that genre. In the front, you'll find some of the collaborations that we've done with the artists, like here is Autograph. And then a bunch of different styles under that genre. So let's say for example, if we pick up main stage progressive, then there's a few parameters or the user can leave it randomized, but we offer the speed.
[00:09:20] Diaa El All: So the BPM and the key major minor. One thing that we'll be launching, deploying here pretty soon. You'll be able to toggle between key and mood. So for content creator can just select mood, happy, sad, energetic, et cetera. And that's all you really have to do. And you click create. And in a few seconds, It produces a track completely unique to that creation.[00:10:00]
[00:10:10] Dan Blumberg: I love it. Can you share more of what's happening in the background? Like how is this happening?
[00:10:16] Diaa El All: So we took an ethical approach. Coming from the music industry, we were very conscious of copyright infringement, protecting rights holders, and making sure that the quality is at the highest possible. So, one, we took an approach of training our models on music theory rules, and we've never trained on any copyrighted materials.
[00:10:38] Diaa El All: We have acquired anything that Sample has, models has been trained on. It has been trained on data sets that we own at 100%. Unless it's a collaboration with the artists, that's a different, completely different way. So, the way that Soundful produces today, it produces based on music theory rules, along with the infrastructure [00:11:00] of a specific style of a genre, along with different models and algorithms for mixing, mastering, um, arrangements.
[00:11:08] Diaa El All: Side effects. It goes on.
[00:11:11] Dan Blumberg: That sounds really great. First of all, that you're doing it in an ethical way. It sounds really, really hard to get that much material. Was that a big challenge is to get that much material that is not owned by someone else to train the models on?
[00:11:24] Diaa El All: Absolutely. We've acquired from certain libraries.
[00:11:27] Diaa El All: We have, uh, collaborated with. With artists and producers. We got musicians into the studio to sample them in a very specific way. We also have our team that has been developing it in in house. So it's a mix between life, traditional musicians and training on the way that they play an instrument, um, all the way to certain synthetic data that we've created by our models.
[00:11:51] Diaa El All: Uh, with scoring algorithms, et cetera.
[00:11:54] Dan Blumberg: Cool. I, I wanna dig into the tech more, but I wanna keep playing for a second here. What, so if we took this track and [00:12:00] say, let's, let's put yourself back in your old shoes as a ghost producer. You, I dunno, did you like the track that just came out? Did you not like it?
[00:12:06] Dan Blumberg: What, what would you, what would you do to change it?
[00:12:08] Diaa El All: That's a great question. So let's say this one in particular, I liked it, but let's say it wasn't really what I was looking for. I can leave every parameter the same and click create again. It will create. similar in the same style. It will have maybe different leads, different melodies, different core progressions.
[00:12:29] Diaa El All: But it will have a different output.
[00:12:35] Diaa El All: So if you notice here, the intro is a little bit different.
[00:12:40] Dan Blumberg: And just, just like using Dolly to create images, you just click a button and said, try again. Right. And this is, this is one of the like magical, but also kind of infuriating aspects of AI, right? It's like, it's non deterministic, right? You get something new each time, whether you wanted something new or not.
[00:12:55] Dan Blumberg: Can you share more of how you work with that unpredictability and [00:13:00] how it's, you know, I presume a feature, not a bug here.
[00:13:02] Diaa El All: Yeah, absolutely. So if a user saves a track, first and foremost, it actually locks in the metadata on the background, so it locks in the metadata where it doesn't, we will never replicate the melody, the core progression, the baseline, so it's not going to infringe on anything else that any other user creates from.
[00:13:21] Diaa El All: Now, also, if a user is creating a track like what we've done in the first one and skipped it and just click create again, well, we take that as data signals to inform our models. That this track wasn't selected and when was it actually click create again? So which part did the user stop in the track?
[00:13:40] Diaa El All: What were the sounds that were playing into this arrangement? What was the arrangement? And then from there, we tried to change it again slightly for the user. And also the most important piece is that every click and every generation here, it is exclusive and unique to you. So not two users will get the same [00:14:00] exact track.
[00:14:01] Diaa El All: Second, you're making sure that it's also not infringing on any copyrighted materials out there, because we haven't trained on any copyrighted materials. And third, the licensing structure is that it's yours. You can monetize on it. You can keep it, you can use it and nobody else will be having the same record.
[00:14:19] Dan Blumberg: Yeah. Talk to me about the value of the human behind the scenes here. One of the interesting things with AI is it levels up amateurs to, you know, let's say semi pro ability, right? I'm curious how you approach that or how you see, you know, AI leveling the playing field or does it level the playing field?
[00:14:36] Diaa El All: This is a great question. We are looked at as a little bit different than just like a traditional AI music creation platform. We are a human aided AI music creation platform. What that means is that we leverage the human in the beginning and in the creation and in the developing these different styles and models to mimic the human way of playing an instrument.
[00:14:58] Diaa El All: And also we [00:15:00] craft sounds like real authentic sounds that we're not synthesizing sounds. So for example, the guitar sound that you just heard. That's actually a musician in the studio playing the guitar. And we took those and sample them in a very specific way to keep the quality high enough as an output for the end user.
[00:15:18] Diaa El All: And from there, we also were backed by the industry. Our first rounds came from some of the biggest artists and producers to managers, to executives in the music industry and in the tech world. And when we were developing the platform. We partnered up with these major artists and producers to develop and help us shape those different styles of what is the trend look like, what are the right sounds.
[00:15:47] Diaa El All: And by having that industry insight, we're always staying ahead of the curve and creating, um, the styles for our users. To make sure that they're keeping up with the trends. Now, your question [00:16:00] about really making producers, whether it's, we are making producers, super producers, or we're really after the market that gets me the most excited is that the people that don't know how to produce and they're great singers, they're great songwriters, or they're up and coming producers that don't know how to put a track of fully composed track together to get to an outcome.
[00:16:22] Diaa El All: We help them. Do that. And we make sure that also when we're helping doing that, we're not, we're giving them a simplified license to just be creative and We're opening up the doors for them to be discovered. And that's, what's most exciting personally for me.
[00:16:38] Dan Blumberg: That's awesome. What's been the hardest challenge in building Soundful so far?
[00:16:42] Diaa El All: Oh, that's a good question. I mean, it's the, it's the blood and sweat and tears of every startup, right? Um, like in the beginning, the biggest challenge was when we started developing Soundful and I started going out with the end of 2020, a lot of the industry, [00:17:00] was pushing against AI. Now, ChadGPT, and Midjourney, and Dolly has changed that drastically for us, um, last year, and we've seen that rise in the wave.
[00:17:12] Diaa El All: But, it was very against, like, the industry was like, some of them were, this is gonna take our job, We don't want it. You're going to end up infringing on our, on our works. That's why we took the approach that we've taken. And that's why we started having like some of the most forward thinking people in the industry started seeing that, no, this is the future and we need to be involved to help shape it and make sure that we're doing it the right way.
[00:17:38] Diaa El All: And that was really the biggest you know, challenge for us to get the acceptance and the validation from the industry. Yeah.
[00:17:45] Dan Blumberg: Yeah. And I think, I think that's an important point that you were working on this before everyone had the aha moment when chat GPT 3.5 came out and it was like, Oh wow. But you were working on this well, well in advance of that, this, this generative AI.
[00:17:58] Dan Blumberg: What about on the technical side? [00:18:00] What's been one of the hardest things as you've just tried to get the output to match what you're trying to, you know,
[00:18:05] Diaa El All: I think that the biggest challenge was for us is scaling up our sound libraries as we use the real sounds rather than just synthetic. sounds because you can always hear if it was created by a machine, you can always tell like this guitar sounds like a little bit synthetic.
[00:18:22] Diaa El All: And we were always like after the best output of quality that can trick the human ear to believe it was done by human rather than done by a machine. And the biggest challenge was really building and acquiring these sound libraries. Arguably today, We own the largest one shot samples library in the world that is proprietary to us.
[00:18:43] Diaa El All: We've built it. We've acquired, we got the musician to be sampling every note, whether it's, you know, an open pluck, a close pluck, um, chords together. And it's always been also, we're only using what it's called a one shot samples. So like a [00:19:00] kick drum is sampled once a Cino on a piano sampled once a strum, a chord strum on the guitar from there.
[00:19:07] Diaa El All: we trained the model of how it was actually sampled and the way that they will strum the guitar and it will take these individual sounds and layering them together the same way that a musician would do it on a guitar. And that's the biggest really advantage for sample rather than just doing, you know, training on a full track.
[00:19:26] Diaa El All: Like full track output and trying to synthesize and create derivative work off of it. We're a little bit different. We're an additive rather than creating derivative.
[00:19:35] Dan Blumberg: That sounds like a very patient guitar player to, to do all that. Yeah. Months. I was going to ask, You're already sort of answering it. How do you differentiate from the many, many other tools that I'm sure are popping up that help you create music using AI?
[00:19:50] Diaa El All: So there's two sides of, of the industry. There is, we sit at a very unique place in the middle, actually. Um, there are certain AI tools out there that are [00:20:00] using loops and they're just taking some loops and layering them together. The problem with that is that you end up with variation problems because you're using full loops, like, uh, whether it's in.
[00:20:11] Diaa El All: eight bar, 16 bar, 32 bar, that loop is going to keep happening over and over and over again, maybe in different settings of the tracks, but it's going to end up sound
[00:20:21] Dan Blumberg: a little boring.
[00:20:22] Diaa El All: Yeah. And repetitive. And, and also you can't really, from a licensing standpoint, you can't really give the copyrights to that person because that loop can be a melody loop and it can be used in a different user's creation.
[00:20:34] Diaa El All: So that's where it gets a little bit. The other side is the large language models, where you just right now, they're just using YouTube and Spotify. Just go and scrub the internet and train on copyrighted works. And by doing that, The machine is not understanding actually the granular inputs and it's just [00:21:00] creating derivative work of that.
[00:21:01] Diaa El All: So it's not understanding what the music theory behind this, why was the chord progression played that way with the melody, with the baseline and the sounds that were selected. So, you know, legal battle aside, because that's a whole different. conversation. But for us, that's why we're, you know, in, sit in a space where we're using those one shot samples.
[00:21:22] Diaa El All: We're building these models to mimic the human way of playing a specific instrument. So we're creating new works and being additive. We're doing a composition of like the chords and the melody from the ground up using the music theory rules, rather than just understanding that hit track and just creating something similar.
[00:21:41] Dan Blumberg: That's very cool. Let's fast forward to the future a little bit. What is something in this realm that sounds very much like science fiction today, but that you think is going to be totally commonplace in a couple of years?
[00:21:55] Diaa El All: A lot of the voice cloning models. They're starting to get better and better and better.[00:22:00]
[00:22:00] Diaa El All: And it's less like I've heard a demo from a company a few weeks ago that they don't have a deal yet with the labels. And that's where the big problem is like the labels and the publishers and everybody's kind of like gunning for the tech world. And that's why we sit in the middle like We are, you know, we looked at as the good guys and we have partnerships with two of the three majors today.
[00:22:24] Diaa El All: But, um, with the voice cloning, that's something that I really feel that it's going to be a big revolution moving forward because don't just think about layering the vocals on the music. But also, for example, for radio, like ad libs or any of that stuff, you can just go rather than wanting like a Nicki Minaj, tell you, Hey, you're listening to Kiss FM, for example.
[00:22:49] Diaa El All: And it's very hard to get an artist to do that. That will be democratized. And that will be easy enough to do, which will create a lot of more creative opportunities. work and derivative work from it, [00:23:00] which is very similar to what we've done with the artists collabs that, you know, if you want to go and half cascade, use a cascade song on your podcast or your vlog or for your commercial, you're going to pay a lot of money for it.
[00:23:13] Diaa El All: And most likely you're not going to get the clearance unless you're somebody very big, right. But we are shaping that in a way of like, no. You're leaving a lot of money on the table. We can create and help you scale yourself to really go into this part of the industry that you haven't been tackling before, because just from a bandwidth standpoint, or from just like, it's not, you know, a hundred dollars, a thousand dollars is not just exciting for them.
[00:23:38] Dan Blumberg: One of the panelists that you were with at South by Southwest where I met you said that he had heard a version of air quotes Adele singing Purple Rain and, uh, and that it would be a hit immediately. It was not, it was not actually Adele and she didn't actually mean, but I'd love to hear it. And, but I, I, it, it opens up all sorts of, you've heard it.
[00:23:54] Dan Blumberg: So
[00:23:54] Diaa El All: yeah, I, I got to hear it and I was like, that does sound like Adele, [00:24:00] but it's still a little bit. Synthetic. It's not there yet. So within whether that is six months or 12 months, I really believe that there's going to be a huge revolution of like the vocal capabilities and what you can do with voice cloning, we're working on a few products and features within Soundful where you're going to be able to upload just an acapella and you select a style like what we've done here and click create Soundful deconstructs that acapella understands the key to BPM and also section analysis.
[00:24:30] Diaa El All: So where's the verse course and it will build music around them so we can create derivative works for these artists. Let's say, for example, a post Malone is releasing an album. We can create a lo fi version, an EDM version, another pop version, hip hop version.
[00:24:48] Dan Blumberg: Stepping aside from music making, there's a lot of fear about AI right now from, you know, all sorts of professions, all sorts of, you know, people just graduating college.
[00:24:57] Dan Blumberg: What do they need to learn? Um, Are there one or two things that [00:25:00] come to mind as like, you should really be playing with this. You should learn how this aspect of it works. That'll really set you up for, for success.
[00:25:07] Diaa El All: Absolutely. Like I, I really, I am a big believer, you know, I mean, I run Soundful as an AI company, so I am a big believer in AI, but most importantly, I'm a big believer in ethical AI technology development, but also in the same way, AI is not the answer for everything.
[00:25:26] Diaa El All: Whether you're a student or an up and coming, whether you're finance or a musician or a content creators, there are tools and I really believe that it's not AI that will replace jobs. It's the person that is leveraging the technology and make it useful for them. This is who's going to be the threat to you.
[00:25:44] Diaa El All: It's at the end of the day, it's the human, because a lot of the models, whether it's in any industry, it's, you know, specifically for music, the output is it can be perfect. Right? But it's not the perfection that makes it relate to other human, it's the [00:26:00] imperfection that I would add to it that will make it unique to you.
[00:26:04] Diaa El All: Maybe I will move a melody, you know, like a piece of a note, completely out of, you know, out of the norm, and I'm breaking the rules of music theory. But that's what human is, you know, is there for. It will make it relate more for you. Oh, that was like, Weirdly interesting.
[00:26:21] Dan Blumberg: Yep. And yeah, and you have to know the rules for you can break them too, which is, which is a, which is a funny thing.
[00:26:25] Diaa El All: Exactly.
[00:26:27] Dan Blumberg: Awesome. I love it. Thank you so much for taking time. This is fascinating stuff. I could go on for a very long time, but maybe let's leave it there. Thank you.
[00:26:33] Diaa El All: Of course. My pleasure. Thank you for having me.
[00:26:36] Dan Blumberg: That's Diaa El All, the founder and CEO of Soundful. And this is CRAFTED., a show about great products and the people who make them. If you liked today's episode about technology and music, and enjoyed geeking out on music theory and machine learning, you'll also love the very first episode of CRAFTED., which features Chris Ladd explaining how he built ChordBank and other apps for guitar players.[00:27:00]
[00:27:00] Dan Blumberg: You can find that episode at the very beginning, and of the CRAFTED. podcast feed in your favorite podcast app.
[00:27:07] Dan Blumberg: CRAFTED. is produced in partnership with Docker, which helps developers build, share, run, and verify applications anywhere without environment confirmation or management. Docker suite of development tools, services, and automations accelerate the delivery of secure applications. Learn more at docker. com.
[00:27:25] Dan Blumberg: Special thanks to Artium, where I launched CRAFTED.. Artium is a next generation software development consultancy that combines elite human craftsmanship and artificial intelligence. See how Artium can help you build your future at artium. ai.
[00:27:39] Dan Blumberg: And CRAFTED. is produced by Modern Product Minds, where I advise companies on product discovery, growth, and experimentation.
[00:27:45] Dan Blumberg: Learn more and sign up for the CRAFTED. Newsletter at modernproductminds. com. I'm Dan Blumberg, and it would mean a lot if you would share CRAFTED. with a friend.
[00:27:54] Diaa El All: That would be amazing.
[00:27:55] Dan Blumberg: [00:28:00] Yeah.