Late Transmissions

Late Transmissions: How AI (and good software engineering) surprised and delighted me.

Catalyst

The catalyst was an innocuous New Year's message we received from a friend. It was a short, fantastical AI-generated video that was uncanny and oddly charming. Watching it, we started wondering about the soundtrack. Was the music AI-generated too? Or was it stock audio stitched in afterward? We never quite answered the question, but it started us down the research path where we discovered tools that can be used for song generation.

We already knew that AI tools for music existed. A family member had even created a full album of country songs using them. We found it impressive, it didn’t spark any urgency in us to try it ourselves. That changed when we realized that some of these tools weren’t just about generating music from scratch. They allowed you to upload your own audio.

That’s when a light bulb went on. Somewhere in cloud storage sat a folder I rarely opened, filled with forgotten work: rough recordings with and without vocals, incomplete takes, old .wav files paired with scanned lyric sheets. Songs that had been written quickly, recorded imperfectly, and then set aside. The digital artifacts were from 30 years ago and suddenly had more currency.

The question wasn’t whether AI could write new songs. It was simpler and more personal: what would happen if I fed these old original-composition songs back into a new system? Would anything recognizable come back? Would it be interesting? Or would it flatten everything into something generic?

Experiment

After a bit of fiddling around, I tried my first test using Suno AI. The results were immediately pleasing and brought a smile to our faces. The tool didn’t feel like it was replacing the songs so much as re-voicing them in interesting ways. (In fact, in Suno it’s called a cover.) With some guidance in prompts and settings, the output came back remarkably close to what I remember wanting to do thirty years ago but couldn’t quite execute at the time. It was thrilling not because it was perfect, but because it felt like I was completing something left unfinished.

I think of these new versions as drafts that are an order of magnitude better than the originals. They’re not finished songs, and perhaps I don’t really want them to be. They aren’t finished because the AI process can introduce noise and artifacts into the final output. These can be removed in remastering Suno output (stems). We didn't go that far for our first attempt because we didn’t know how. And even with the noise and artifacts in our final songs, we were happy.

My preference for these songs is for real people to perform them someday. Or is that idea anachronistic? At least now these drafts can properly communicate the idea, the mood, the structure I had in mind. That, to me, is magic: not automation, but translation.

Expression

During the rediscovery and reimagining of these songs, I came around to the idea that songs function for me much like blog posts. They are audio blog posts. (Yeah, I know that much of the world has moved on from longer-form writing like blogs, but I haven’t.) Blog posts—and now songs—are formats in which I can express my thoughts or feelings.

Some of the messages from thirty years ago survived surprisingly well. Others needed a small amount of tuning, though I resisted changing the lyrics more than necessary for this first try. I wanted to hear what that earlier voice was actually saying, not rewrite it from the comfort of hindsight.

I was surprised at how many of the themes addressed in the songs are still relevant to me today: my voice, my sense of freedom, and ability to deal with modern life. In short, timeless themes.

In that sense, reinterpreting of these songs isn’t about reviving the past so much as finishing a sentence or blog post.

Release

Using AI in the process wasn’t about polishing or correcting the past, but about filtering it. It acted like a lens, separating signal from static, letting certain ideas come through more clearly than they ever had before. The result isn’t a set of finished songs so much as clearer drafts, and that feels right for now. Releasing them now isn’t a performance or a bid for attention; it’s simply a form of publication. It’s much easier to send a link than a batch of files. Publishing lets these transmissions complete their journey so I can move on to the next phase.

This post—and the album that shares its name—are both called Late Transmissions (Spotify, Apple, YouTube). The title felt unavoidable. These songs were written long ago, partially broadcast, and then abandoned mid-signal. They’re late not because they were delayed on purpose, but because they took a long route to their current form.