Audio supports out-of-core processing. We can create a file-backed Audio object using
au = Audio[File[...]]. Any filters applied to it will create a new file-backed Audio, and processing will happen without loading all the data into memory at the same time.
Is there a way to create our own audio filters that can act on file-backed Audio objects, either in pure Mathematica or in C/LibraryLink? The typical filter would need access to an interval $[t-\Delta t, t+\Delta t]$ to be able to compute the amplitude at $t$ (similar to
Note: I sent the same question to Wolfram Support today, however, at this point I have little hope for a documented solution. Audio is still very new, and I expect lots of improvements and additions in the future. I am posting the question here on SE in the hope that there may exist a simple undocumented way that I haven't discovered yet.