Mubert: finding product-market fit, building a text-to-music tool, and venturing into new fields
Photos provided by Paul Zgordan
During the spring of 2022, Mubert stopped part of its processes – primarily development and marketing – because of the war that unraveled between Russia and Ukraine. The company left only one-third of the whole team as many had to leave their homes, relocate, and some had to save their families. Mubert almost stopped working at all because everyone was so stressed, and even considered selling out. Fast forward to 2023, the startup now has a new CEO, shifted focus to its two main products, and is ready to reach new heights.

"It was hard for us to continue but we managed to do that," Paul Zgordan, now Mubert's CEO, recalls. Deciding to participate in Wallifornia Music Tech was a turning point for the company: the team studied different topics about the music industry, had great advisors from top music companies like Warner Music, and through that understood what next steps they could take to pivot. It was also a big personal step for Paul, as he got involved in more business processes – before, as the Chief Content Officer and music director, he primarily managed the creative and production side of projects.
With a fresh perspective, the company began working on realizing ideas with web3 companies: one of the projects was done with NFT artists on generative soundtracks, which is set to release this year. "We made a very complex algorithm to generate tracks for NFTs with different tiers of [tokens]," the CEO explains. "We were thinking that web3 is very interesting for us and almost pivoted to an NFT platform, but realized that it's better for us to stay in our generative music tech product. [While] we want to partner more with web3 companies, we aren't a web3 company [ourself]."

Mubert did, however, jumped on another rising trend and developed its own text-to-music generator. Paul spent a lot of his free time generating images with different neural networks and noticed that a lot of AI artists doing similar things were asking for AI soundtracks to accompany their art. Some were recommending Mubert in the different chats and communities. It made a lot of sense for the company to go in that direction as it already had its own base of metadata (compared to other startups that didn't have the datasets for it since music is more complex). Launching it as a prototype on Google Colab, it didn't take long for users to start testing the product, and it quickly became one of Mubert's core features.
"Music is more complex – you have to make a lot of work to get these large datasets for music generation, but we already have them from our musicians and sound designers who were uploading all of these sounds to our servers and tagging it. We have a complex tagging system, which was [originally] created for Mubert Render for search optimization, but can be used for text-to-music generation."
This is the first prototype, so it's largely based on samples. While there aren't any neural networks used for raw audio synthesis, the team is planning to expand the features of this product and add some NN to scale. The text-to-music generation has also attracted a lot of attention, especially from investors and industry specialists. "It brought us more audience, we doubled our revenue, [and] we got five times more users in two months," Zgordan says, demonstrating how the startup is growing its user base. The product is also very simple to use: in order to get a soundtrack, all you need to do is type in a prompt and it almost instantly pumps out music.

The CEO shares that while Mubert has been in the generative content space for five years, the team has started getting a lot of attention from the industry and seeing validation of its ideas only now. Moreover, the startup is well-positioned because it works with artists and musicians, paying them for the sounds they submit. "It's a very ethical way to get data," he adds. "[Musicians also] know how we use their sounds." A lot of the time, artists aren't credited for the way their art is fed into neural network training; they aren't asked for permission. To further support musicians, Mubert is developing a revenue-sharing system that will allow a wider range of artists to get a new revenue stream with AI-curated content. For instance, if you created some music samples, you can upload them to the company's servers and get some money from its user subscriptions. "I think that is more future-proof than different methods of how developers get content for datasets," Paul concludes.

Another big turn for the startup was Alexey Kochetkov stepping down as CEO and Paul becoming the new one. As Alex was getting more involved with his second company – making digital art, NFTs, and other creative media – transferring his role to then-COO was a logical step. Being the founder, Kochetkov still takes an active part in the development of the company, serving as the strategic adviser. In this transition, Paul wanted to introduce changes in the operational activities and product development. A lot of those ideas were born during Wallifornia, which helped the startup understand its product-market fit.

If before Mubert had six main products (a mobile app for personalized listening experiences, Mubert Render for content creators, for venues, and more) that were all targeted for different markets, Wallifornia made the team realize that its main product was the tech itself. Instead of fighting for users, it became evident that it's best to focus on one of the main products – Mubert API (the other being Mubert Render) – and offer it to those same competitors. The other product, Mubert Render, is also a primary focus for the team and, in some ways, acted as a demo of all the possibilities API could offer.
"Now, we see all of our products as demos for various marketplaces: [they] can win from this situation and add API to their software," Paul says, sharing that they already have some pilots with music stocks. The company can generate tracks for the stocks from its dataset, while also using their samples to make custom datasets catered to their purposes too.

A key project for Mubert in 2022 has been the partnership with Anghami and Visa. Before the 2022 World Cup, Anghami contacted the startup and asked for B2B access to its API. Of course, the team inquired what exactly would the streaming service want to make with it, and Mubert began working together on this collaboration. This project was designed to generate musical cheers for football fans – users can pick a country they're cheering for, and the tech generates a unique track created just for them. "We set up our generator with some custom engine for this product and generated more than 200,000 songs for the fans," Paul conveys. Anghami recorded a variety of vocals from different countries, and those vocals were added to Mubert's generative content. This collaboration marks the first time any company has generated so many songs with the help of AI.
"This year, we added six albums to Anghami exclusively to our generative content, [with] tracks for different activities. We are going to add more content to Anghami and test this [idea of] distributing music through [various] DSP platforms."
For 2023, Mubert is bullish on fine-tuning its products, which includes improving its text-to-music engine. The CEO conveys that the team has a big roadmap of features and improvements in plan to make the output even more relevant. The startup is currently fundraising to garner more resources to allow for this development. "This fundraising round will help us hire more specialists, DSP developers, and data scientists – the main challenge is to keep this growth running," he adds. "We are pioneers in this field and we want to keep it going."

Mubert is also developing features for the metaverse with partner companies like Sensorium. "We [do] want to reach others as well and [are] discussing collaborations with other musical metaverses," Zgordan says. There are also plans to venture into the gaming industry, as the CEO mentions that the startup is open to finding new partners in the space. "Activision pub PR – generative engine for online games to get music for their game services – was a good sign for us that game developers are also thinking about generative content," he elucidates. "We'll be glad to work with indie developers that are very experimental and test different hypotheses with adaptive sound."

In comparison to other fields, games have an adaptive music system based on particular surroundings, the plot, and other factors you choose. This can be tricky for musicians to create these types of soundtracks as it requires special software and additional skills. As Paul says, with Mubert's tech, it would only require a couple of API requests. "The success of our competitor Splash in Roblox shows that the gaming industry is ready," he concludes.
Made on