New v0.3 update:
Added the ability to filter a corpus/dataset (using fluid.datasetquery~). I still find the syntax for that object really terse, but wrapping it up in layers of abstraction has made it much easier to use (e.g. filter loudness > -35 or time_centroid == 50, etc…).
For now it’s a one-query-only thing due to how it’s abstracted, but I may figure out a way to make it still readable to do multiple queries in the next update.
Also do all the scaling post-filtering, which is quite nice as otherwise the descriptor space gets pretty fucked.
Also completely revamped the playback stuff with a more “full featured” playback sampler (start/length, attack/hold + curves, speed, etc…) and added in some “realtime” descriptor analysis (JIT stuff based on the fftsettings). I’m wanting to implement some mosaicking options in future updates, so that’s setting the groundwork for that eventually.