CCB Magazine

In 2018 they were participants of the Invest in Creativity Investors Lab Berlin, today they are already spoiled for success: Cyanite is a startup that analyzes and visualizes songs in terms of their emotional profile. The program works with artificial intelligence (AI) and is intended to help actors in film, advertising or games with music selection. The three founders work together from Mannheim and Berlin, we spoke with two of them.
 

INTERVIEW BORIS MESSING  

CCB Magazine: Hi Jakob, Hi Markus, in 2018 the three of you went public with Groovecat, a video social media app, and a year later you finally founded the music AI startup Cyanite. Tell me, who are you guys and how did you meet?

Jakob Höflich: We met in 2014 in the master's program Music and Creative Industries at the Pop Academy in Mannheim. During our studies, we had already come up with a few ideas. It started with the idea for an app that shows you what kind of music the people around you are listening to. For example, you're sitting on the bus and think, oh, what kind of music is the person in front of me listening to, and you look at the app, see that they're listening to what you like, and start a conversation. That was the basic idea at the beginning. Then came Groovecat, our first app for music moments. With Groovecat, you can combine music with a video that you record to capture a special moment and share it with others.

CCB Magazine: Your new startup Cyanite uses AI to analyze and recommend music. Please explain how that works? What’s the scope of application?

Jakob Höflich: Cyanite is about finding the one perfect song for your purposes. Cyanite's algorithm therefore compares the emotional profiles of thousands and thousands of songs to specify the search. Let's say you're looking for a song like "In the air tonight" by Phil Collins, but with a female voice, guitar instead of synthesizer, and a slightly calmer mood with a melancholic touch. This is exactly the song you are looking for. Using various parameters - genre, mood, tempo, timbre, etc. - you can now narrow down this search with Cyanite, refine it and lead it to the target. The algorithm filters a selection of possible matching songs from your music database and the bottom line is to facilitate professional work with music.

There will always be those full professionals like Quentin Tarantino who don't need music recommendations from an AI. But there is also an extremely large target group for whom a program like Cyanite is a blessing

Jakob Höflich. Foto: Maxim Abrossimow

CCB Magazine: Although you haven't been around for very long, you've already had some success. Who is your target group? Who is your product aimed at?

Markus Schwarzer: We sell our technology to various companies such as SWR, RTL or BPM Supreme from the USA. But also DJs and music publishers are customers of ours. Cyanite works with the companies via an API interface. In doing so, the AI analyzes their entire music database so that songs can be better classified and categorized. This allows ad producers to find more targeted songs for their commercials, radio hosts and DJs to find songs for their sets, or music publishers to catalog their media libraries more efficiently. No one has hundreds of thousands of songs in their head that they can compare with each other - Cyanite can do that.

CCB Magazine: Ok. But how did the radio and TV stations organize their media libraries before? It seems to have worked before, too.

Markus Schwarzer: Manual. Manual keywording. Some have employed an armada of interns for this, while others, who value a tidy music database, have hired professional musicologists for this. In the long run, however, this is expensive and not very efficient. With us, on the other hand, you only pay a monthly subscription depending on the size of the database, and the AI continuously improves through so-called Deep Learning. For example, a few weeks ago we analyzed 900,000 songs for a major client in just 15 hours.

CCB Magazine: To put it bluntly, why do you need an AI to find a suitable song for a commercial or other purpose? Isn't the producer's intuition or the art director's knowledge of music enough? And anyway, where's the fun in choosing a song yourself? I mean, Quentin Tarantino would never think of letting an AI dictate the songs for his films, for instance.

Markus Schwarzer: Well, Quentin Tarantino only makes a film every few years. Thousands of other films are produced during this time. And it's not so much about movies; advertising productions, for example, need much more and faster music content, and advertising on social media channels like YouTube also needs to be constantly fed with music. A database analyzed by AI guarantees a faster and, above all, more targeted selection, so Cyanite professionalizes the production process. This is a useful tool, especially for people who don't have such a strong understanding of music as Tarantino. Music can simply be better assessed this way.

Music publishers used to hire an armada of interns to sort through their music databases. Recently, Cyanite analyzed 900,000 songs for a client in just 15 hours

Markus Schwarzer. Foto: Maxim Abrossimow

Jakob Höflich: To add to that. There will always be professionals like Tarantino who don't need such an AI for music recommendations. But there is also an extremely large target group for whom such an AI can be just right. For example, if you just type in the hashtag "Finding music for my video" or similar hashtags on Twitter for fun, you become aware of the mass of people who have no expertise and no overview. They think to themselves, there are a thousand different platforms, there's so much music out there, how am I supposed to find that one song that fits my video? I have a deadline, it has to be fast. Using Cyanite here can be very helpful. For example, in combination with Spotify.

CCB Magazine: How can Cyanite be used with Spotify?

Jakob Höflich: We create our solutions mainly for companies for a fee. However, there is also the possibility to use a free version of Cyanite and have smaller music databases analyzed by the AI. Many DJs, composers, musicians or producers use this for themselves. And in this context, music from the Spotify library can be analyzed, or even your Spotify mix of the year.

CCB Magazine: You've already mentioned that Cyanite's AI is self-improving through Deep Learning. That sounds kind of creepy, like the AI is a living thing on its own. Can you even understand how the AI improves?

Markus Schwarzer: It's actually difficult to see behind it. Joshua Weikert, our third co-founder and Head of Technology, could certainly explain it better. How the AI arrives at its judgments can only be guessed at. In this context, we also speak of a black box. The task of one of our data scientists, who also conducts research in this area, is to make the AI's decisions and judgments more comprehensible. So that you can see, for example, ah, because of this frequency, the AI has now assigned the song to this mood. At the end of the day, the process of Deep Learning is similar to human experience gathering. So the whole thing is actually not that scary.

CCB Magazine: Cyanite not only analyzes the songs for similarities, but also visualizes them by fanning out their structure in terms of various parameters such as mood, tempo or pitch. Why does one need this visualization, don't one have a natural sense for songs? I mean, whether a song is sad or happy, fast or slow, I can recognize that myself.

Markus Schwarzer: Sure, if you're the producer of a song, you know it inside out. But our customers have music databases with 100,000 or 200,000 songs, of which listening through even just a few would take a lot of time. By tagging the songs, i.e. breaking down their structure in terms of the various parameters, you can see at a glance how the song is structured. Here the beat changes, there the key changes, etc. Or if, for example, you need a song that can't have percussion at this point or that point, or has to have a crescendo, you can see that in the tagged song. And looking at a tag like that is quick, listening to the song all the way through takes time. So you can quickly assess what is suitable for the purpose and what is not. Cyanite can also be used to analyze individual sections of a song and then search for similar songs. This expands the possibilities of finding the right song or the right part of a song.

CCB Magazine: What about competition, is there any?

Jakob Höflich: There are several companies like Musiio, AIMS or MusiMap that are in the same field as us. Just like us, they got started a few years ago. The AI technology we work with hasn't been around that long. That's why startups like ours are only now coming onto the market.

CCB Magazine: Speaking of the market. What are your plans for the future of Cyanite?

Markus Schwarzer: I can't imagine any player in the music industry not using this technology in five years. AI simply promises a very clear process optimization, so everyone has to follow suit. Many are already using it, and many of the big players are in the process of deciding in what form they want to integrate such AI systems. Some are doing research on it themselves. And some are coming to us.