Hi!

Does anybody know why mp3 encoded audio sample (disabled all filtering) has different spectrum than initial wave file, in high frequencies especially. Lame gradually cuts off the spectra although it should be the same with -k mode.

Thank's!

One word: Psychoacoustic

http://en.wikipedia.org/wiki/Psychoacoustic

because LAME is lossy.... pretty simple explaination

LAME or any other mp3 encoders will still cut off most of the frequencies above 16khz even if you disable all filters

it's the nature of mp3 not to store many info above 16kHz ....most people can't even hear it!

Along with psychoacoustics, it's partially to do with the Absolute Threshold of Hearing (ATH). Lossy encoders can mangle high and low frequency data more than the mid-range because hearing is less sensitive at the extremes of the audible range.
http://en.wikipedia.org/wiki/Absolute_threshold_of_hearing

Ya. I thought about it. Thank's!

One detail: different versions of LAME do this "job" differently. For example, 3.93.1 removes very little. Instead, 3.96 cuts off almost all frequencies above 16 kHz.

Is this differences in psychoacoustic model used?

Yes

Now I'm more interested in this "Absolute threshold of pain" that's listed in the psychoacoustics article. Seems odd that it doesn't exist yet.

it exists
http://en.wikipedia.org/wiki/Threshold_of_hearing