I've been making my own audio codec (nothing meant for public distribution; I just want to learn how to do it) and I have a question regarding masking and quantization. Consider the following audio spectrum:

where the black line is the actual power spectrum, and the red line is the calculated mask.

I know that at point A you can completely throw out the frequency since it's below the masking curve, but what about point B? How does the masking curve regulate the proper quantization? Assuming the spectrum is at 50dB and the mask is at 45dB, does that really mean you can add another 45dB of quantization noise to the signal? I just don't know how to interpret the output.
Any clarifications would be of great help.