Hello,
I was pointed to this article claiming, that MP3 is cutting off high frequencies. I didn't believe it, so I did my own test. I ripped a WAV from CD and analyzed it. Then I encoded and decoded the track using LAMEDrop (3.97b2, quality 80, ~200 kbps VBR) and analyzed it again. I found out, that the spectrum is almost identical to the original sample up to frequency 16 kHz. Then it rapidly drops and there is nothing more over 19 kHz. Why is LAME cutting the frequency this way? I believe a lot of people are able to hear frequencies over 16 kHz. Unfortunately I'm not one of them.

Here are the graphs I got:




Vlada

a good codec by a frequency response that accurately follows the source material to at least 16kHz, no reduction in spectral content below 16kHz and the inclusion of spectral content above 16kHz.

source: lame dev Gabriel's website

the guy on that article says



what to say?
bull sh..

Short answer: Yes, it does, as you have found out.

Longer answer:

Why does it do so? Because the MP3 data stream has limitations on how much data to store and in the way that data is stored.

As you can see, it follows very faithfully up to 16Khz, and then there is a drop at 16khz.
This frequency is not casual. It is where the last Scalefactor Band starts ( SFB21 ). Trying to maintain this scalefactor down, allows for more bits to be used in the lower bands, where artifacts could be heard easier.

At last, i would like to point out that the graphic you show is not good for any quality measurement.
A spectrogram could serve a bit more, but it wouldn't be good either. Basically, your graphic has a sum of all the data to a plot. It seems as if all content over 16Khz is attenuated, but it is not.
What happens is that selectively it gets cut under some thresholds, and thus, that data cannot sum up to the original signal.
Over 19Khz (18700), there is really a filter being applied. If you use a higher quality, the filter will increase, and more bits will be used.