Hello everyone! Today’s post is a bit unusual for my blog — it’s about the problem of removing vocals from a song. Actually, this problem can be solved very easily with the help of a neural network (I used this one), and the backstory of the issue is tied to nostalgia.

Since the distant days of my childhood, which fell on the 1980s in the USSR, I’ve always loved music. One of the brightest memories of that time is the amazing electronic music (surprisingly, there was quite a lot of interesting electronic music in the USSR — but that’s a topic for another time). This music was featured on the record “Rhythmic Gymnastics”, released in 1985 with an absolutely fantastic print run for the USSR. By the way, here’s the album cover:

I had almost forgotten about that era, but today, completely by chance, a track from this album popped up in my YouTube feed. It instantly triggered a wave of nostalgia. I couldn’t resist — I found and downloaded the entire album and started listening to it. But then I suddenly realized that, despite the beautiful electronic music, the voice of the instructor (which I’m honestly not interested in, though it’s quite pleasant 🙂) was distracting.

This led me to a question: how can you remove the vocals from a song in a way that leaves the music intact and still enjoyable? I remember back in the days of dual-cassette tape recorders (the wild ’90s!) and the emergence of mixing consoles, we managed, through trial and error, to minimize the influence of vocals on music. But, as you can imagine, the final quality was awful. 😅

However, progress marches on, and — drumroll! — the Moises neural network blew me away. (There are some usage limitations, but… they can easily be bypassed with disposable email addresses. 😉) See for yourself:

Original track:

Processed track:

I’m sure you’ll enjoy the result as much as I did! 🙂

And with that, let me wrap up. Thank you for your attention!
If you have any questions, feel free to write to me at email or on Telegram.