[Guy Sherman]
Would the approach to use a sample-rate converter to
essentially interpolate
samples, then do the processing, and then sample back down?
The principle is indeed the same, and you could use a converter
library for this purpose. However, those converters are designed to
work over a continuous range of samplerate ratios whereas the ratio
chosen for oversampling is usually a fixed integer because this
presents ample opportunity for optimisation. The interpolation
filters in both cases are usually windowed sinc FIR (much like the
Lanczos kernel in image resampling).
How does that work for live streams of data?
As you intuit: you sample up, process, then sample back down, ending
up with one output sample for every input sample.
IIrc,
http://quitte.de/dsp/caps.html contains at least two oversampled
plugins and comes with source code.
Cheers, Tim