This is awesome, I've dreamed about something like this for a long time.
Cool that it's done in Clojure too :)
I've had some really good results with sample-based algorithmic composition, so I'm really looking forward to this:
"In the near future, Alda’s scope will be expanded to include sounds synthesized from basic waveforms, samples loaded from sound files, and perhaps other forms of synthesis."
It's easy to see how a programming language can be a good fit for notating certain kinds of music and allowing playback of it. Seems to me like the challenge is to go beyond lining things up neatly to a grid and introducing any kind of "looseness" to the music.
Edit: Then again, people do great stuff with trackers, so I'm sure a programming language can be a great fit for some kinds of music.
Though you may already know, aligning notes to exact subdivisions of the beat is called quantization.
https://en.wikipedia.org/wiki/Quantization_%28music%29
Quantized music feels noticeably less organic and lively, to me at least. Audio sequencers like Linux's Hydrogen give you the capability to mess with "lead" and "lag", but it must be done individually for each note and so is fairly tedious.
I've used the Python module pydub along with a random number generator to set samples slightly ajar, but, while it helps, a purely random approach doesn't capture how human performers implement lead and lag.
I absolutely recommend pydub as an easy way to manipulate sound files.