If you plan on making a lossy file with a lowpass of, say, 16KHz, is it advisable to resample the 44.1KHz input to 32KHz, since the Nyquist Theorem states that a 32KHz sample rate is all that a 16KHz lowpassed signal needs? Resampling will lead to additional space savings. However, resampling will produce quantisation errors (audible or not) since 32KHz and 44.1KHz are not evenly divisible.

If you use a tool like SSRC, perhaps the resampling issue is mitigated. However, what if you used "--resample 32000" in LAME? If you are seeking transparency (aps) I'd assume you'd stick with 44.1KHz but if you are making lossy files for a portable is resampling an advisable method to reduce bitrate?

In general, absolutely not. Lossy encoders are smart enough not to allocate space for the unused high frequencies and they are often less well tuned for non-CD sampling rates.

The block sizes in mp3 are in fixed numbers of samplea.

So, at 32kHz the block length is longer (in time) so pre-echo is worse.

Cheers,
David.

LAME automatically resamples to 32KHz if you use -V 7 or -V 8. What I found interesting is that if you do "-V 7 --resample 44100" the files are about 10% larger even though the higher sampling does not capture any additional frequencies because the lowpass is already below 16KHz. So the bitrate savings is coming from somewhere...it sounds like the block sizes are the root cause.

Principally it does. However try once this line below with Lame 3.94b (or above):
Lame.exe -V7 --lowpass 15.5 input.wav output.mp3
It doesn't matter if you take '-V8' or '-V7' by the way.
I also had expected an Mp3 at 32kHz at that time. - Regrettable that my knowledge about Lame inside and Mr. Nyquist himself isn't very large-scaled...


Maybe i do not understand well now...
But isn't it plausible that a vbr encoder allocates more bits each second when treating 32000 sample per second compared with a encoding of a 44.1 kHz sampling rate, specially when both times lowpass filter, noise-shaping, etc. are identical!?


I know no other reason why not to downsample exept this one.
Specially at files there downsampling belong the theory gets attractive and if the resulting mp3 file size have to be below a certain value the user must decide between pre-echo (sharpness) or the encreased effect of a too low kbps-rate.

Some time ago I started a similar topic. I found out then that I'm able to ABX the resampled version of a song, but am not able to ABX the lowpassed one. I'm still curious why it is like that.

Interesting test you propose there.

This make me curious how much higher frequencies generally are in my favorite tracks and how far i'm able to hear them.
So i have applied a hipass-fliter of 16kHz on them...

... there were as good as no perceivable sounds left - at least for my ears.

Longer block time with lower sample rates shouldn't be a problem as long as the encoder can use short blocks effectively.