Chris Johnson
June 30th 03, 11:40 PM
In article >,
"David Morgan \(MAMS\)" > wrote:
> Average RMS amplitude is determined by the entire file content *selected*.
> Some sections of the song may be higher, some parts lower... it is not a
> constant, therefore the term "average".
>
> You would use "Peak" normalization to accomplish the goal you describe
> above. Why? There could be specific segments of the program that
> are actually much higher in RMS than the "average" would indicate. It
> is these sections that would suffer the most traumatic damage.
This is one hell of a good point, particularly for someone like me
who hacks on DSP software a lot. When I read this whole topic it's as a
closet sympathizer with Myke: why shouldn't you be able to seamlessly
cast all possible music into any given 'mold' that you happen to like?
I write batch-processing MacOS 8 audio mastering software (no, not
OSX- no, not PC...) and one of the tricks I've done well with is
metering. I render the song visually across sort of a 'strip chart',
with peaks represented as black speckles, and RMS loudness over time as
a gray region that goes higher on the chart and is rendered darker, the
closer it is to -0dB RMS.
As such, I've seen exactly what David is describing, _visually_.
Bass notes produce very high RMS when they occur. Certain vocal
sounds, like sustained vowel sounds well-recorded, can produce very high
RMS. The funny thing is, when music is very dynamic, these
characteristics can express themselves and the RMS is all over the map
and varying wildly, but when music is squashed and peak-limited, not
only does most of the RMS come up, but these moments of high RMS come
down. You don't see brief bursts of supersuperhigh RMS loudness, instead
you just get an unvarying blare. Bass in particular loses its capacity
to be unexpectedly loud and present, and the vocal resonances also stop
being any different from the background sound, and sort of merge with
the music.
Again, this is not based on listening opinions (though that IS what
happens), I'm reading this stuff off a computer-generated, calibrated
chart of RMS/peak over time. There most certainly are isolated sounds
that produce unusually high RMS, and that is most certainly lost when
you peak limit.
Chris Johnson
"David Morgan \(MAMS\)" > wrote:
> Average RMS amplitude is determined by the entire file content *selected*.
> Some sections of the song may be higher, some parts lower... it is not a
> constant, therefore the term "average".
>
> You would use "Peak" normalization to accomplish the goal you describe
> above. Why? There could be specific segments of the program that
> are actually much higher in RMS than the "average" would indicate. It
> is these sections that would suffer the most traumatic damage.
This is one hell of a good point, particularly for someone like me
who hacks on DSP software a lot. When I read this whole topic it's as a
closet sympathizer with Myke: why shouldn't you be able to seamlessly
cast all possible music into any given 'mold' that you happen to like?
I write batch-processing MacOS 8 audio mastering software (no, not
OSX- no, not PC...) and one of the tricks I've done well with is
metering. I render the song visually across sort of a 'strip chart',
with peaks represented as black speckles, and RMS loudness over time as
a gray region that goes higher on the chart and is rendered darker, the
closer it is to -0dB RMS.
As such, I've seen exactly what David is describing, _visually_.
Bass notes produce very high RMS when they occur. Certain vocal
sounds, like sustained vowel sounds well-recorded, can produce very high
RMS. The funny thing is, when music is very dynamic, these
characteristics can express themselves and the RMS is all over the map
and varying wildly, but when music is squashed and peak-limited, not
only does most of the RMS come up, but these moments of high RMS come
down. You don't see brief bursts of supersuperhigh RMS loudness, instead
you just get an unvarying blare. Bass in particular loses its capacity
to be unexpectedly loud and present, and the vocal resonances also stop
being any different from the background sound, and sort of merge with
the music.
Again, this is not based on listening opinions (though that IS what
happens), I'm reading this stuff off a computer-generated, calibrated
chart of RMS/peak over time. There most certainly are isolated sounds
that produce unusually high RMS, and that is most certainly lost when
you peak limit.
Chris Johnson