On 2024-07-29, I stumbled upon a demonic rendition of "Jingle Bells". It turns out a computer sang it.
The technology is the Festival Speech Synthesis System, a 1997 invention by Alan Black, Paul Taylor and Richard Caley at the University of Edinburgh. Though not designed to sing, the feature was implemented "as a student project by Dominic Mazzoni".
Surprisingly enough, the previous stable release (v2.5) was 2017, and more surprisingly it runs on my machine (though not without a few tweaks). It segfaults sometimes, but works.
You know what else was born in 1997 and remastered in 2017? OK Computer.
Naturally, the first track that comes to mind is Fitter Happier, which is just spoken voice. But I need it to sing.
So I yanked some midis from the internet and wrote a Python script to convert it, along with lyrics, to an XML file that Festival could sing out.
I tried Exit Music first, but because I do not have unlimited time, only the climax.
Then I stayed up late to make the entirety of Paranoid Android. The robot seems unable to comprehend the concept of BPM, and always sings too fast. Not only that, but it has trouble keeping on time. Once the audio is exported, it's more manual labor aligning them on the DAW timeline.