An ATH chart shows that frequency Y is "inaudible" below amplitude X.

In a given WAV, frequency Y's amplitude is X-2dB, therefore, frequency Y
is discarded and does not become a part of the final MP3.

However, after "normalizing" the WAV by a factor of +3dB, frequency Y's
amplitude now becomes X+1dB, therefore, frequency Y is retained and
becomes a part of the final MP3.

AH.... I think I see the problem with your understanding of frequency
masking (as it was originally called).


The threshold of audibility is not a defacto amplitude. It is the amplitude
of a one frequency as compared to another frequency.

The codec works based on the concept that one loud tone will mask our
perception of another quieter tone which very close in frequency to the
louder tone.

If you have a tone of kHz at -5dB and another tone of 1.001KHz at -20dB,
that second tone may be removed from the signal since it is unlikely that
anyone could distinguish the presence of the quieter tone.

If the kHz tone goes away and the 1.001KHz tone is still at -20dB, the codec
will restore the 1.001KHz tone to the signal. as it will now become audible
since it is no longer masked by a louder, nearby tone.

If, however the second tone is several octaves away from the first, say
200Hz at -20dB, it will still be audible even though it is much quieter.
Since there is not another, louder tone nearby to mask it, will be audible.


- FLINT