Disclaimer: this post is mainly for those familiar with the work of legendary Soviet rock musicians. But even without understanding the words, you can still see here how songs are already being created with the help of neural networks. Of course, without the ability to compare with the original works, the perception of these songs will be quite different.
After Ozzy’s death, a wonderful picture like one above began circulating on the internet. I think if Ozzy had known who Tsoi was, he would have appreciated such a posthumous joke about himself. The text over it is says:
— And where is Tsoi?
— But Tsoi lives!
Because for my generation, the phrase “Tsoi Lives” was never a meme. Viktor Tsoi is a legend — ours, our own.
And in memory of Viktor, new works keep coming out. His already famous song “Peremen!” became one of the anthems of the 2020 protests in Belarus. Viktor’s son brought together former band members and created a remarkable project with modern arrangements of the songs, over which his father’s cleaned-up voice was laid.
And with the development of artificial intelligence, people have gone even further. I once stumbled across one such song, listened to it, and forgot about it. But the other day I was shown another one — and suddenly the puzzle came together.
There is a user, AiRushV, who became interested in neural networks and started experimenting with them on musical material. Essentially, he would “feed” the neural network the works of a famous musician and ask it to create a new song that this well-known but long-departed artist might have written. The results he then turns into music videos — also with the help of neural networks.
At the same time, he does not hide that this is the work of neural networks. Right from the start he writes — this was made by AI, this is not the voice of the performer. Of course, to get a good result, one also needs to craft the prompts correctly, so there is some share of “authorship” from AiRushV. But he does not try to claim the glory for himself.
But why am I even writing about this? Because after listening to several of the resulting songs, I was genuinely impressed. And specifically by the songs created from the material of all the albums of the band Kino. Yes, you realize this is not Viktor Tsoi’s voice, though it resembles it. But the lyrics, the style of performance, the music — though a little different — I can quite believe that this could have been made by Viktor and his band, had he lived to our time. And I know Tsoi’s work very well: as a teenager I listened to all of Kino’s albums to death, I bought the posthumous Black Album on vinyl at a kiosk on the very first day of release, practically waiting by it for the delivery. That is why I decided to share a few of the tracks that I liked most.
The first is the song “Tishina” (“Silence”), created about three months ago. This is the very one I was shown yesterday. Viktor against the backdrop of modern-day St. Petersburg, yet as if he were someone who had left his homeland because of the war.
“And the Sun Will Rise” is built on a slightly different narrative, but in the video created by AiRushV it also pays tribute to all the rockers who did not live to see our days.
The third song, “I Am Alive,” directly touches on the subject of war. In the video, the neural network shows certain soldiers without insignia. But from some of the details of the events shown, to me they look more like Ukrainian defenders — even if the neural network does not depict this explicitly. That is how I will perceive them.
Beyond Tsoi himself, the creator also experiments with the work of other legends. “Skeleton and the Jester” is a tribute to Gorshok and Korol i Shut — specifically their early period with its dark, fairy-tale ballads. The voice, I would say, sounds less like Gorshok’s own and more like that of his younger brother, the frontman of Kukryniksy. But that’s tricky, since the brothers’ voices and manner of singing are extremely similar. As for the style, though — yes, this really does sound very much like Korol i Shut.
But the attempt at reproducing Grazhdanskaya Oborona and Yegor Letov is probably the least close to the original. The song itself turned out good, but the voice doesn’t really resemble Letov’s at all. Although, I admit, I was never a big fan of his — maybe at some point he did sing that way.
But overall, I’m very impressed by what can be done now. And this does not take anything away from the original musicians. These generated songs would not exist if there were nothing to train the neural network on. Yet in music, a new genre is clearly being born. Before, we knew covers as re-performances of other people’s songs, and now we have songs “in the style of.”
