Does anyone know what the difference, when using lossy compression, between appling a Low Pass Filter at ~16kHz to a 44.1kHz piece of audio compared to resampling the audio to 32kHz?

If you are encoding at 128k, for example, is there a theoretical benefit from resampling because you have to throw away less bits to reach the desired bit rate?

Just curious.

-Iain

assuming we're talking mp3 here, there's a definite advantage to resampling - the sfb21 bloat problem exists even if there's no frequency content in that band (i know, stoopid).

assuming there's no horrid hardware resampling problems in sound cards, you'll get a more efficient encode with 32khz rather than 44.1+LPF

mp3 aside, there shouldn't be a difference, if nyquist is correct and the LPF and resample are implemented in a sane way (downsampling usually involves a LPF before the downsample anyway).

If we do not consider the mp3 case (sfb21) resampling to 32kHz would be more efficient because the frequency resolution would then be more precise for the codec.

I never knew that a universally excepted theory in signal processing could be called into question in such a casual manner. It's like saying "if pythagorus is correct" when talking about right angled triangles.

Is there any doubt that Nyquist's Sampling Theorem is correct?

@ cabbagerat:

of course i know that. no problems with putting things into context? hmm. i think i have a tendency to inject a certain irony into what i say. this could be misinterpreted easily.

if the soundcard isn't b0rk, the algos are all good, etc, then there will be no difference in the LPF and the resampled file according to Nyquist. this is the same as saying "if he's right there'll be no difference"

Would temporal resolution be compromised? I mean, would pathological transients be more audibly smeared in a 32kHz MP3 than in a 44.1KHz MP3? A quick castanets.wav test might answer this for me, but I don't have any audio equipment available to me right now.

Yes: the more you decrease the sampling rate, the more time resolution is decreased.

Our beloved natural audio mpeg codecs are using block sizes expressed in a given number of time samples, but unfortunately this size is the same whatever the sampling frequency is.

Sorry about that. I think I have been reading DiyAudio too long where this kind of thing is questioned as a matter of course.

Was this a conscious design decision on the part of the MPEG or just a side effect of some other compromise? If it was a design decision then do you know what their motivation was?

It is a design choice, in order to keep memory requirements constant over different sampling rates.

i completely understand that... it's good that HA is always on the lookout for voodoo physics and such.

I think in the case of AAC, downsampling from 48 -> 32 khz will not contribute that much to pre-echo audibility..

From (256/48000 5 milisec to (256/32000 8 milisec will still result in inaudible pre-echo..