The Long Way to the Songs in My Mind

I am not a musician.

Music is one of those things where we quietly assume that having clear ideas and being able to bring them to life are the same thing. I don't think they are. I’ve loved music for as long as I can remember, but I never had the kind of physical intuition for instruments that some people seem to be born with. When I watched friends play guitar in high school, it felt like watching someone speak a language I understood perfectly but could not pronounce.

I did eventually learn to play, after a few hundred and thousand of hours of experimenting, I learned a little. Enough to know that I was always an order of magnitude behind where my ear already was. The music in my head was coherent. The music coming out of my hands wasn’t.

What I really liked was not primarily playing music, but asking questions about it. What would this song sound like if the melody was like this ? If it were slower? If the harmony moved earlier? If the melody were carried by something brittle instead of warm? Most listeners hear a song as a finished object. I always heard it as one point in a large space of possible versions.

In 2012, I discovered MuseScore and started playing around trying to compose little songs.

This felt like a breakthrough. I could finally put notes down without my fingers getting in the way. Unfortunately, what came out sounded awful. Way worse than what I was hearing in my mind. At first this was discouraging, but over time I realized it was informative. There were two obvious problems.

The first was orchestration. I didn’t know which instruments to use to express an idea. I could hear the contour of something, but not its clothing. The second problem was more mundane: the instruments themselves were bad. Cheap sound fonts are unforgiving. They make everything sound like a mistake.

I always listened widely to music. Baroque, rock, punk, Eastern European traditional music, Latin music, musicals, pop, instrumental, modern classical. This helped my taste and knowing what I'd like to hear but not my confidence. The more I listened, the harder it became to decide what a given melody was supposed to be. The same melodic idea could plausibly live in several genres, and choosing one too early felt like closing doors.

Lyrics created a different problem. I enjoyed immensely writing them, but I couldn’t and still cannot sing with my voice. That's painful when you love music. Worse, I couldn’t even imagine how the words would sound with a voice. Lyrics without a voice are like dialogue without actors. Technically complete, emotionally inert.

I published a few tracks anyway. MIDI renderings. Some light mixing in Audacity. They received no attention, which was not surprising. No one could hear the thing I was hearing. Potential is invisible if you can’t point at it.

The next shift came with Synthesizer V Pro in 2023.

Suddenly I could add voices (used Solaria, heavily processed in Audacity). Not my voice, but a voice. This helped more than I expected. Even imperfect singing carries information that instruments can’t. My music improved, but it still didn’t feel even remotely finished. It sounded like a good sketch of a painting someone else might someday complete.

I tried to skip that step by working with musicians directly. This was expensive and mostly disappointing but a great lesson and reason for asking many questions I still don't have answers to. Paying people to record your music does not automatically turn them into collaborators. Without shared context or investment, the results were often technically correct and emotionally wrong or all together wrong. And I learned something else: people are very reluctant to collaborate with unknown 'composers' unless money is involved.

Then I discovered that Suno could do covers.

This was the third phase, and it changed my process a lot. By generating covers at the maximum similarity allowed to my original MIDI and Synthesizer V versions, I could hear my own songs realized with some competence, much closer to reality to my ear. Not perfection, but plausibility. More importantly, I could hear them across styles. What sounded awkward as a chamber piece might suddenly make sense as a rock song. What felt thin as pop might work as something closer to folk or modern classical.

The control is imperfect, sometimes frustratingly so. You don’t get to specify every instrument or articulation. But the tradeoff is speed. I could finally iterate at a much faster pace and learn (the learning before was extremely slow and would have taken me decades to do the same thing). Instead of wondering whether a song might work in a certain style, I could just listen and having to imagine variations I could pursue.

At this point it became clear that the AI wasn’t replacing necessarily musicians in my process. It was replacing the part of the process that had been blocking my musical sense from expressing ideas. Suno Covers let me surface ideas that were already there but inaccessible. This made me very happy, like an impossible dream coming true, if only for the reason that I could now hear the melodies I had in my head now realized.

People often frame tools like this as shortcuts. To me that’s not quite right. To me they are bridges. They connect intention to outcome for people whose problem was never imagination, but bringing these ideas to life somehow.

This makes me suspect that the long-term effect on the music industry won’t be fewer musicians, but more composers. Or perhaps more accurately, more people who think compositionally. People who hear structure, contrast, and narrative, but don’t necessarily have the hands, lungs, or years of training to express them directly initially (e.g. as Demos).

Whereas we have millions of musicians, one could count only a handful of non-musicians music composers (e.g. Irving Berlin, Diane Warren and more recently Max Martin). Irving Berlin lived in Tin Pan Alley Era, when songwriters were in some ways the main product, Diane Warren mastered the “power ballad formula”, and Max Martin got his start with Cheiron Studios. How does a composer (not a singer-songwriter) gets to record their songs nowadays ?

Historically, we’ve been comfortable with tools that lower the barrier to writing or drawing, but uneasy when they do the same for music. That may be because music feels more physical. But if you strip that away, composition is just organized listening.

AI tools like Suno Covers don’t create taste. They don’t create curiosity. They don’t decide what’s worth keeping. They amplify whatever is already there. For someone without musical sense, that’s not enough. For someone with it, it can be the difference between silence and something they can build on.

I still prefer working with real musicians one day, truly collaborating - not just paying for a service. Human voices and instruments carry friction and surprise that no model quite reproduces. But preference is not dependency. The songs exist regardless. They just take different shapes depending on what options are available.

In that sense, this isn’t really about AI. It’s about removing barriers between ideas and their expression. And when that happens, you don’t get fewer songs from humans. You get a much larger surface area for talent to emerge.

Which is usually how progress actually works.