How does buffer size affect resampling?

[anstahc]

I was under the impression that buffer size was irrelevant when resampling. I just ran a null test at 48 samples VS. 2048 samples. Big difference, and most audible during note transients. Any knowledge on this?

[fishmonkey]

are you absolutely sure that each render is identical? note that any effects or instruments that are not completely locked to the tempo may produce slightly different results in different renders. if you are not sure if this is the case, try making multiple renders with the same buffer size. do those renders null?

[anstahc]

Well I'm saying I'm sure that they are not identical because they aren't null. I thought they would be identical. No effects added. Just to be clear I resampled MIDI to audio track, not rendered mixdown or freeze track. I just tried it with Operator and the same result. If time sync was an issue the entire pattern would be not null, not just the beginning of the attack. But the null test reveals audible sounds during the note attack and then silence during the sustains.

I also tried this with 5 different instruments from Trilian, Kontakt, etc. I resampled all at 48 and rendered the complete mix, then did the same again at 2048 buffer. I then A/B'd both and the 2048 mix was significantly superior. Way more clarity and way more up front than 48 which sounded like it was pushed back. Clearly it is a quality issue that is throwing the null test and not timing/latency. I'm all sorts of confused because I thought buffer size wasn't a factor in the resampling signal path.

[fishmonkey]

Well I'm saying I'm sure that they are not identical because they aren't null. I thought they would be identical. No effects added. Just to be clear I resampled MIDI to audio track, not rendered mixdown or freeze track. I just tried it with Operator and the same result. If time sync was an issue the entire pattern would be not null, not just the beginning of the attack. But the null test reveals audible sounds during the note attack and then silence during the sustains.

i'm not talking about overall time sync, but small differences in the sound, e.g. due to free-running oscillators that are out of phase.

if you make multiple renders at the same buffer size, do those renders null?

also, presumably you can play the original unrendered tracks at 48 samples without glitching?

[anstahc]

I was using sound libraries so there shouldn't have been any discrepancies. I ran it one more time this morning and everything was null. Weird anomaly or I screwed up somewhere. Anyway thanks for trying to help, I think I made both of us dumber in the process. Does buffer size matter in any resampling situations, like say you have an FX chain or a Max For Live device, or is it always irrelevant?

[fishmonkey]

I was using sound libraries so there shouldn't have been any discrepancies.

just because you are using presets from sound libraries doesn't guarantee that those particular sounds are always identical though.

anyway, back on the buffer front, theoretically it is possible that there is some interaction between audio buffer size and MIDI timing to plugins. have you asked Ableton support yet?