Chibi Nappa
New member
I agree that everybody should be aware of intersample peaks and know how to avoid clipping them, but I wasn't talking about clipping. I was talking about masking.
A bit about masking and mp3's:
Mp3's look at the whole sound. They decide what a human will hear and what a human won't hear. Then the mp3 throws a lot of bits to the part of the wave we can hear while throwing very few bits to what we can't hear.
As an example, if I'm talking in a low voice in a quiet room people will hear that. If I'm talking in a low voice in a quiet room and then a lawnmower wheels in and parks right next to me, people most likely won't be aware of my voice at all.
An uncompressed recording keeps all of the information that makes up my voice. An mp3 would look at the parts of the wav that make up my voice and reproduce it in very low resolution while accurately reproducing the parts of the wav that make up the lawn mower.
That is amplitude and frequency masking.
The other case is time domain masking. Say I'm in a quiet room and I drop a pencil. You'd hear it. Now say I'm in a quiet room and I shoot a gun before I drop the pencil. The gun shot and the pencil drop do not occur at the same time. If the pencil drop is close enough in time to the gun shot (the pencil has to come second), you will not hear it.
That is time domain masking. It depends on the second sound being much quieter than the first. In this case, the mp3 would produce the wav very accurately for the gun shot, and then the pencil drop would be very lo-fi.
Hot masters remove much of the loud/quiet difference on a micro level. Whatever comes right after a drum hit is no where near as quiet as it would be otherwise, and therefore can not be masked nearly as well.
So that's what I'm talking about. Keep your louds loud and your softs soft on a micro level and the mp3 will encode better.
This has nothing to do with macro dynamics. Quiet verse/loud chorus will not affect time domain masking. Only dynamics that occur in milliseconds.
A bit about masking and mp3's:
Mp3's look at the whole sound. They decide what a human will hear and what a human won't hear. Then the mp3 throws a lot of bits to the part of the wave we can hear while throwing very few bits to what we can't hear.
As an example, if I'm talking in a low voice in a quiet room people will hear that. If I'm talking in a low voice in a quiet room and then a lawnmower wheels in and parks right next to me, people most likely won't be aware of my voice at all.
An uncompressed recording keeps all of the information that makes up my voice. An mp3 would look at the parts of the wav that make up my voice and reproduce it in very low resolution while accurately reproducing the parts of the wav that make up the lawn mower.
That is amplitude and frequency masking.
The other case is time domain masking. Say I'm in a quiet room and I drop a pencil. You'd hear it. Now say I'm in a quiet room and I shoot a gun before I drop the pencil. The gun shot and the pencil drop do not occur at the same time. If the pencil drop is close enough in time to the gun shot (the pencil has to come second), you will not hear it.
That is time domain masking. It depends on the second sound being much quieter than the first. In this case, the mp3 would produce the wav very accurately for the gun shot, and then the pencil drop would be very lo-fi.
Hot masters remove much of the loud/quiet difference on a micro level. Whatever comes right after a drum hit is no where near as quiet as it would be otherwise, and therefore can not be masked nearly as well.
So that's what I'm talking about. Keep your louds loud and your softs soft on a micro level and the mp3 will encode better.
This has nothing to do with macro dynamics. Quiet verse/loud chorus will not affect time domain masking. Only dynamics that occur in milliseconds.
