Assuming that I could record at 88.2 kHz or 96 kHz... I may some times in the future want to convert to 44.1 kHz or 48 kHz, but I cannot predict which of these I will want and it could even be both...
I read at least two theories:
A) Resampling will be better if the original is a multiple of two in regards to target (so 88.2 kHz -> 44.1 kHz, 96 kHz -> 48 kHz).
B) It doesn't matter at all, resampling from 96 kHz to 44.1 kHz is as good because resampling is not such simple mathematics?
Which one of A or B is true ?
Which one of A or B is true ?
B.You can't simply throw-away every-other sample because you can get aliasing. The audio has to go through a low-pass anti-aliasing filter before downsampling. The filtering is more complex than the actual downsampling and the filtering ends-up altering every sample to some extent anyway.
In terms of quality resampling, the answer is B. The best sinc-based filters are within rounding error that the only loss of information is in the stop band — rejecting frequencies above 20KHz or so, which can be configured to be the same regardless of the incoming sample rate.
Real reasons why someone would consider an integer downsampling ratio is either higher performance using IIR filtering (which is not as mathematically sound or flexible as FIR, but handles integer ratios fine), or for complete control of phasing — 2:1 samples with 88.2-to-44.1 instead of 320:147 for 96-to-44.1, that splicing together audio after downsampling will accumulate fraction-of-sample delays. That aside, quality resampling will deliver virtually the same results for the resampled audio regardless of sample rates due to the nature of Fourier transforms, retaining all the frequency energy within the limits of the filters.
Assuming that I could record at 88.2 kHz or 96 kHz... I may some times in the future want to convert to 44.1 kHz or 48 kHz, but I cannot predict which of these I will want and it could even be both...
I read at least two theories:
A) Resampling will be better if the original is a multiple of two in regards to target (so 88.2 kHz -> 44.1 kHz, 96 kHz -> 48 kHz).
B) It doesn't matter at all, resampling from 96 kHz to 44.1 kHz is as good because resampling is not such simple mathematics?
Which one of A or B is true ?
Sounds to me like pov A is a misunderstanding or misapplication of the Nyquist theorem, which says you need twice the sampling rate of the frequency range being measured to accurately represent it.
However, more relevant to the proposition you state is the fact that hardware design does involve tradeoffs, and that results in the selection of specific targeted or native sample rates "currently" based on the standard pro audio sample rate of 96 kHz, or perhaps multiples of that with newer hardware.
Best thing then is to try to capture at your card's native sample rate, and then once you have the signal you can resample to your heart's content in the digital domain without penalty.