Name: Voicing The Future
Uploaded: 2025-07-23
Duration: 14 min 54 s

Kyogu Lee, Co-founder and President of Supertone, is reimagining voice for the age of artificial creativity. He reveals how Supertone’s ethical, AI-powered tools are amplifying artistry and reshaping the future of sound. From multilingual vocals to real-time vocal transformation, his work is opening new possibilities for creators in music, film, and more. Featuring a live tech demo, we'll explore how AI can amplify human expression without replacing it.

Voicing The Future

Conferences

OK No. So thanks everyone for joining my debut as a K-pop star. There will be autographs after this session, but in all seriousness, you know, um. In all seriousness, thanks to Supertone, I was able to be part of this duet with Cindy, a group made of AI-based voice characters. I'm excited to be here with Supertone co-founder and president Ko Go Lee. Artists around the world are using AI to expand their narrative and creative possibilities rather than fearing it will replace them. Singers are working to actually co-create. with voice AI tools and we're going to take an in-depth look at some of the real world tools that are changing creativity and artistry. So let's get started, you know, I want to start by asking like what you guys actually did. So for context, I gave Superone three videos of me singing in my hotel room, Call Me Maybe by Carly Rae Jepsen. Uh, I heard it through the Grapevine by Marvin Gaye, Pink Houses by John Mellencamp. I gave Supertone these videos and you know, voila, thanks to AI, what you heard was the result. But what actually is Supertone doing with with these voices? OK, so let, let me first uh ask you a couple of questions. Uh, do you speak Korean? No. OK. Are you a good singer? I don't think so. OK. So, uh, and that's what we achieved at Superone. So let me tell you, uh, so we developed a foundational model for voice generation and you can think of it as a larger language model like KGBT or loud, but for voice. So, our model can generate incredibly natural and expressive voices, including singing voices using inputs uh such as uh a speech, singing voice, text or melodies. And in this demo we uh transformed the original singer's performance into your voice. We call this uh thing with convergence. So unfortunately we didn't use any of your videos. We just took the audio of your recordings and to add more detail so our model can. Uh, take as an input any speech or sy signal and break it into four different attributes. The first one is a pitch and the loudness, I can whisper, I can speak louder, right? And the timbre, which reflects most of your or anyone's the vocal identity and linguistic content. Of course, it depends on what kind of language you are speaking. So we have two recordings, right, original singer performance and your recordings. So we take them as inputs and we break them into these four attributes I just I just mentioned and we took. Uh, the, uh, the timbre part from your recording and the rest the three, the lingu consonants pitch and loudies from the original performance, and you can mix them together to reconstruct the singing voice which is in your voice. That's what happened, and, and there's, there are some other things that your that that your AI can do and you're gonna show us actually now kind of what what your programs can do to someone's voice. OK, so, uh, imagine you're an artist. And working on your next album. And for this new album, you want something different, you want something new, uh, totally different from your previous albums. So you have, uh, you are ready to explore this bold new idea, but your natural voice doesn't quite match well with your artistic vision. So with this uh voice technology. Uh, we can instantly, uh, transform your voice into other musical personas to bringing new, uh, different musical identities to life. So, uh, let me show you how it works right here live because, uh, listening is believing. So when I speak, Like this The energy does a shit, doesn't it? So you can feel a different kind of presence, a calm, yet very powerful. Now, notice how capital my voice changes to the feminine to. It's much lighter, brighter. And carries a complete uh different kind of energy in the field. I can even sing very high notes that I can with my natural voice. So you know, I, the, the clip we showed earlier was, was me and a duet with the virtual group Cindy H. it's a fully AI generated virtual group, you know, what does it mean to have a fully AI generated pop group? What does it actually mean in terms of how they perform? What music they're putting out. What does this project actually mean? OK, so let me explain. So this product, it was a very bold experiment uh from our side as well. So we started two years ago and we wanted to showcase our like a voice technology, how we can uh empower the, the creators or, or the world uh to unlock their creative potential. So we formed uh this kind of virtual group. And uh which is composed of four female singers. So let's, uh, I mean, assume that you are a producer. So you have this idea of this virtual uh girl group and we have different uh personas, uh, four of them. And uh we have songwriters, we have some music, and we have a production engineers and everything. So you have the songs and everything is ready. So if you're a producer, what would you do next? What, what's gonna be your next step to actually produce music? I mean, they, I mean, you have to let them sing and record the performances and Uh, in, in, in the legacy, uh, media production, you need to, I mean, resort to human, uh, Personnel, I mean, the singers, artists, right? And you need to uh find uh the very good artist that matches your artistic vision. And we, I mean, usually do an audition or casting uh to accomplish that, but in this case, we designed a unique voice, voices for uh each of them. So our technology can uh design or create uh theoretically unlimited number of unique voices that Or never heard before. So for this uh 4 members, we created their voices uniquely and using our singing voice synthesis, uh, model, we put the melodies and lyrics and let the models sing. So with all of these AI models that that can generate kind of these new voices, they are always very impressive, um, they are very impressive about what they can do. They're also a little bit, you know, spooky at times, you know, and there have been, there have been debates about, you know, about voices sounding like celebrities and celebr. being upset about that, but you know how is Supertone navigating this? What, what are the kind of the, the rules of the road that need to be set down involving, you know, voices who owns them, the voices you generate from them? Are you thinking about these ethical questions? Yes, this is a very good question. So, uh, the respect and consent, uh, for the audience rights must always come first. So a Superone. We, uh, we have very clear and transparent licensing frameworks and the uh license approved, uh, the, the uses, uh, from the, the artists or creators themselves. So we never use their vocal identity without explicit and informed consent. That's how we uh build trust between the the the. The artists and the creators, so they, that's why how we can, I mean, uh, accomplish this uh past projects over the past few years, you know, Korea's cultural industry is very strong, you know, K-pop, K-dramas, K music. Um, and you know you're operating K-pop space. You work with hive, you know, what's the interest in Korea for embracing these new AI tools? Are they excited to adopt them? Is there some hesitance? What's the mood like in Korea about? Well, it's, it's, it's both, I mean, like every other country, I mean some people see it as a huge potential for the content creation. New ways of uh exploring the fresh ideas and everything, but the other parties, especially the uh the artists and the creators themselves, uh, some of them uh look at this technology as a very uh threatening because it could um replace the, the human artist, but I don't agree with that because I mean. Uh, it's tool. I mean we, we see the creators and artists as co-creators or collaborators, uh, which means that in, uh, they, uh, we collaborate with them not only in some specific project. But also when we uh develop and refine the technologies so that uh we can reflect their feedback and concerns to better shape our products and uh and the, the tools. You know, I was thinking about, you know, technology and music and the idea of virtual bands. I mean, it's not a new idea. Guerrillas in 2005 is a virtual band. Japan has vocaloid software where you can use to generate songs. Japan also has, you know, their own virtual pop stars. Yes, Avatars in front of a human performer, but still, I mean, there the persona is, is artificial. Um, but this seems to be kind of a step beyond that. Where do you see entertainment developing, whether it's in terms of AI assisted artists or even wholly AI generated artists, where do you see entertainment developing in the next, you know, 5, even 10 years? Well, one thing for sure that uh this before we say that technology along with the other technologies. Uh, it will, uh, democratize the creative process for sure. So the, uh, look at every creator or artist can uh expand with new technology to uh explore and experiment with new ideas. I mean, it can empower, empower them to, uh, I mean, yeah, uh. Expand their boundaries and deepen the connection with their fans and collaborate with other artists as well. So I think it's a uh uh both a an opportunity for the creators and the and the uh uh so that they can, I mean, uh engage uh with audience. More immersively and more interactively and, and how do you see Superton's journey over the next 5 to 10 years? You know, Entertainment will change and there will be more brace these AI tools and artists. How do you want Supertone and its journey to change over the next 5, 10 years? Well, so. We, uh, as I said, over the past years, we have, uh, collaborated with many artists, producers, directors, writers, and all kinds of creators, and they help us, uh, define and reshape our tools. And they are also uh affected by the tools and the technology we have developed so they can try new experiments and everything. So I mean, as, as, as with all the technologies we have, uh, witnessed, uh, so far, I think the AI is is another yet very powerful tool for co-creation and collaboration. So that's where our super uh what I mean, has been. Having and this will be the way that we can still uh explore with the the creators and and and and so when you do this, this kind of voice work I mean who ends up who ends up owning the final product, um, you know, is it, is it the person who contributed their voice is it Supertone kind of how does the. Either like who owns it and how do you think like these thoughts about IP need to change with more AI systems? Yeah, it's like a conventional music production if I uh take an example for music because I mean in music there are many uh right holders, right? The the production company and the Uh, the agencies and the and the singers were performance as well. It's, it's same as this. So it's like, uh, I mean, if we uh contribute to a singing voices uh by uh uh by obtaining the rights for the uh the voice owners, then, uh, the, the singing part belongs to this voice owner or artist or whoever. Um, you know, maybe to help kind of close things up. I mean, obviously this, this works in Korean, it works in the K-pop genre, but you know what's a music genre that you might want to try next? If you could, if you could spend time to get this AI model right, what's another genre of music that you would really want to test this with new genre or existing genre? Well, I'm as a, as a huge fan of the, the 70s, 80s wrong in metal band and yeah, uh, rest in peace of Osborne who passed away. Yeah, so yeah, that's because I mean it's, it's very technically challenging and the growling sound and everything, the vocal idity, that's very challenging. So that's, I mean, as a technology provider I like to, that's where I like to try and experiment with my, I mean our technology. Do your, do your models yet work with other languages other than Korean? Oh yeah, sure, yeah, it can work I mean in any. Well, I think this has been a great conversation. It was really great to, to see this work for myself again to see, to see my voice change into this K-pop song. Um, it really was a remarkable experience. I really want to thank you, Kyogu and Supertone team for walking us through this and showing us how this all works. Please thank Kogu and the Supertone team for their work. Thank you very much. So yeah.