GPTsongs




hardware:   Apple MacBook M4 Air, Eurorack modular system/hardware, Roland VT-4 vocal effects processor
software:   Apple Logic Pro X, Max/MSP, RTcmix

Part of the reason I do 'simple' AI in these pieces (i.e. I generally just give a single prompt and accept the output) is that I haven't really invested the time in learning how to use the more powerful features of AI assistants like Claude or Codex. My skilled-coding friends like Doug Scott and Joshua Mailman use them quite a bit, and they claim it really helps speed the code-development process. Even son Daniel uses Claude to help with his neuroscience experiment data-analysis.

I'm also using the most basic, free versions of these tools. I think I can get access to the more fully-featured versions of some of them through Columbia, but I'm not at the point yet where I think that's necessary,

I'm generally happy with the way these came out, except for GPTsong3. The rotating chord-sequence for that one come as a demo preset with the Rossum Panharmonium module, and I didn't do as much DSP on it as I might have. I guess I was thinking about maintaining the directness of the sonic output from the tools I chose to use. After putting it away for a while and listening again, though, it's kind of growing on me. I did major remixes several times to wind up with the current GPTsong3 version. I could probably have done more, but it was time to move on.

As I said in the main page for these, I decided to use the voice-samples as sonic objects, not so much as texts to be 'set'. I also wanted to explpre ways of looping external effects devices into my Logic mixing application. I used my Roland VT-4 vocal effects processor and my Rossum Panharmonium Spectral Resynthesizer module to accomplish this. Fun sounds!

Piece website:

Brad Garton
August, 2026