First AI Steps

My initial research into AI music led me to the platform Mureka.ai. There, you can create a song with just a few words that sounds incredibly good. I experimented a bit and tested some of my old song lyrics. The resulting songs were decent. "Decent" is an understatement, because it takes many attempts before a song is truly acceptable. Mureka now also has a studio function that allows for better editing, but to get a really good song, numerous attempts are necessary. You rarely get that lucky break.

However, I come back to one of my most frequently used phrases in this whole project: "Yes. But no." Yes, it sounds good. You could probably even release it. But no, it's not my music, it's just my lyrics. No matter what I try, uploading my own old songs, it adopts the style but not the melodies, the structure, etc. They aren't bad songs, but they're simply not my bad songs. So the search continues.

My search leads me through:

Kits – primarily for AI voices, which I'll probably use later in the process.

Udio – neither better nor worse than Mureka.

Ace Studio – a MIDI-to-voice and instrument converter, which I'll likely use later as well.

to Suno. Suno is considered an industry standard. Whether it truly is, I can't say. But Suno is the only generator I've found that can non-destructively rework my songs. Suno retains the ideas, melodies, etc., and simply expands or re-records them. You can use sliders to adjust how closely it adheres to the original song, how closely it follows an additional style description (which needs to be quite detailed), and how "creative" Suno becomes during the creation process.Das für mich extrem faszinierende dabei ist - und das ist auch der Punkt, an dem ich mich entschließe, das Ganze als Projekt weiterzuführen und diese Webseite zu erstellen - wie sehr ich die Arbeit mit Suno genieße. Suno bringt eigene Ideen ein. Gut Ideen ist vielleicht etwas viel gesagt, es ist eine KI und die bringt ein, was Millionen von anderen bereits getan haben. Aber für mich ist es so, als würde ich wieder mit andern Musikern zusammen arbeiten. Man wirft etwas ein, ein anderer interpretiert das und dann entwickelt man eine Idee weiter. Ich verwende also zunächst Steinberg SpectraLayers Pro 12 um den Gesang aus alten Aufnahmen zu entfernen. Dieses Tool besitze ich bereits seit längerem und nutze es, um aus Songs, das Schlagzeug zu entfernen und als Play Along zu nutzen. Suno bekommt von mir den Text und interpretiert die Gesangslinie komplett eigenständig. Die Ergebnisse sind sehr interessant, zeigen aber auch, dass nicht jeder Text geeignet ist. Manche Textstrukturen, in denen die bspw. der Rythmus der Worte variiert wird kann die KI nicht gut umsetzen.

The decision to re-record all the old songs I wrote with my band 30 years ago in this way was quickly made. For many years, I've longed for the freedom of making music. My time is too limited due to family and work. The AI ​​helps me because it's available exactly when I have time. It's where I am, and I don't have to travel anywhere. It masters every instrument perfectly—unlike me. But first, I need to make a plan for how I'm going to create the songs.