On 4/21/19, Nicolas George <george@nsup.org> wrote:
I apperciate the efforts to reply more than half a line.
Paul B Mahol (12019-04-21):
It is not kind of resampling. Resampling is specific and belongs to separate library.
There is no doubt it IS a kind of resampling: it is in the name. The question is whether is it specific enough to warrant a separate interface.
https://dspguru.com/dsp/faqs/multirate/resampling/ Resampling involves interpolation. If I do resampling with aresample and resampling with factor 2 from 44100 to 88200 I can see there is still some spectrum data in highest frequencies.
The oversampling is another thing, and doing only upsampling/downsampling via libswresample is fragile/incomplete, as you need some kind of lowpass. (be it IIR, FIR or simple FFT) after upsample and before downsample.
There are various way of doing a resampling with various properties. This is just another one.
If you haven't noticed, this is just upsampling/downsampling without lowpass. And that is not resampling in any way.
That is why it best fit into lavfi because it is just one block of processing with multiple solutions.
Making it a separate filter is the easy solution, but not the good one. It is reporting all the choices to later or the end user. That makes for easy quick code now, but terrible use later.
That is just your opinion now, you need to provide technical terms to support your statements.
Also how that could be exported? It does not belong there.
Options. We already have them, just add a few. And in that case, it becomes accessible for all filters.
Another option to aresample filter would just confuse users. Two options which would set rate with completely different outcome.