Composition Convolution
I've been fiddling around with techniques to digitally sing the songs I compose!
One of the reasons I started posting my blogs here was because I assumed I’d get notifications from the mobile app whenever someone commented on one. Apparently not, though! So I only just noticed the comments on the previous post.
Anyway, I decided to use the past few days of 2025 to rest a bit in preparation for starting properly on the first Monday of the year, which is today.
I didn’t do nothing while ‘resting’, though. I composed a couple of pieces of music!
I want to make some progress on Dreamons, and for ages I’ve been meaning to compose theme songs for all of the Dreamon characters - with lyrics that describe their worldview - similar to this one I made for Spryad ages ago:
I composed similar songs for the whole cast of Memody: Sindrel Song over the course of the first month or two of 2019 (wow, time flies), and I feel that was what gave me the motivation to stick with it through to completion since it felt like such a big, important part of the project was out of the way early on.
I’ve been struggling to even start on songs for the other dreamons, though, so as a way of sort of practising, I composed some for some other characters I’ve included in non-game things (I suppose they’d count as ‘my OCs’, though it always feels weird thinking of them that way).
I often use ‘lyrics’ for my music, but they’re typically left silent because I can’t sing (I assume; I’ve never even dared to try). Years ago, I found some obscure program that could synthesise very-robotic-sounding ‘singing’ from midi data like in this example:
(That’s Collie’s Theme (V2) from 2021, apparently?? Time flies… “Our world’s rough, tough, broken / But there’s angels!" etc.)
I find that amusing, though it’s not something I’d use for a final result.
It seems that generating singing like this should be one of the many things AI is capable of these days, so I looked that up, but a lot of what I found generated the melody and/or words too (of course, since the people using it aren’t inclined typically inclined towards being creative), or… well, I just didn’t like the idea of using AI for the same sorts of reasons that others oppose it.
I did consult ChatGPT - an AI - about alternatives, though, and it suggested Vocaloid and things adjacent to that, which I’ve been aware of for years without really understanding exactly what they are. I still don’t fully understand. They seem like a mostly Japanese (and Korean/Chinese?) thing, and the English-speaking communities I found seemed full of anime fans speaking about ‘voicebanks’ as if they were their waifus or something?
Apparently there’s a program called OpenUTAU (plain UTAU seems to be an outdated precursor) which can synthesise singing, though the exe has no licence and throws a ‘Windows Defender blocked this file’ error when I try to run it. I’ve investigated enough to know it’s probably safe, probably, but I’ve yet to push through the reluctance to take that chance (and I’m not tech-savvy enough to set up a virtual machine or whatever).
I did however solve an issue this morning that I’ve had for like two decades, kind of thanks to ChatGPT, so that’s something?
I compose in Sibelius - and have been doing since maybe a few months after starting, in my teens, a lifetime ago - and originally used the basic midi soundset; all of the music in MARDEK used that. I eventually got frustrated by its limitations, and learned about soundfonts, but as far as I was aware, I couldn’t use them in Sibelius, and had to export my compositions from Sibelius as a midi that I’d import into another program (called SynthFont, I think?) and apply the soundfonts there. That’s what I did for projects like Clarence’s Big Chance, Alora Fane: Creation, Miasmon, etc.
A while back, I looked into Sibelius alternatives - since I’m very aware I’m neglecting important parts of music production that other composers see as crucial - but never got past fumbling ‘I can’t immediately figure this out’ frustration with a DAW.
I learned about some more modern notation-based programs which were similar to what I was used to, though, and downloaded MuseScore, though it didn’t feel like enough of an improvement over Sibelius to switch over completely. Even though it allowed me to easily apply soundfont files.
This morning I tried (the free version of) Dorico - another modern Sibelius alternative - which required me to create a Steinberg account, download the Steinberg Download Manager to download the Steinberg Library Manager to install Dorico which I had to download the Steinberg Activation Manager to activate, which forced me to sign up for a newsletter (declining aborted the install)… so that was fun.
Especially since I likely won’t even use it! Both it and MuseScore lack a feature I use a lot in Sibelius: the ability to export videos. It seems like such a straightforward thing, but MuseScore has some convoluted method where you upload to their website, then ‘publish to YouTube’, and can only do that once a day (which is very much less than ideal for me since I publish to video several times over the course of working on a piece), and - as far as I can tell - Dorico has no ability to export to video at all.
So that was a waste of a morning.
But! Eventually I discovered - after talking to ChatGPT, buggering around in menus, and installing some other thing called sforzando - that Sibelius CAN assign other soundfonts to instruments!! Which is like solving a mystery after two decades of (half-hearted) wondering.
So maybe that made all the effort worth it??
The main reason I was even motivated to bother with that though was because I wanted one of the songs I’ve been composing to use a specific ‘doot doot’ voice instrument (rather than a more choral ‘aah’ sort of one), the same one in this piece that I composed for - but didn’t include in - Taming Dreams:
(That’s Elwyen’s Theme. Back then, I wasn’t writing full lyrics, but instead used a character’s ‘motto’ (I think that’s the term I went with?) for the main motif; here that was “You can’t not notice a flame so hot it burns blue”, which starts at 0:14.)
So now I can use that instrument in Sibelius… though it might all be for naught anyway if I end up replacing the voice instrument in any songs I make with something synthesized by OpenUTAU or similar!
I’ll need to experiment with that and see.
Either way, experimenting with all of this after sticking with the familiar for so long - and having at the back of my mind the vague intention to change things up at some point - has at felt like some kind of a step forward, even if it’s a tiny one. So that’s not nothing.
I also meant to address the issues with my website by basically recoding the whole thing using more modern techniques… after taking the time to learn those techniques. I intended to start on that today, though I did the music stuff instead since I’d already spent the past couple of days on that and wanted to finish what I’d started.
I actually looked into it a few days ago, though - again with the help of ChatGPT - and it looks likely that I’ll end up using some PHP thing called Laravel (though I’m bracing myself to get scoffed at for my ignorance). Getting that to work will require downloading a bunch of stuff, though, about which I felt reluctant… so maybe now that I’ve downloaded this stuff for music composition, I’ll be less wary about pushing through that when I get back to it. Hopefully that’ll be sometime in the coming days.