|
|
|
|
In article ,
says...
Doing some from CDex, EAC, and Razorlame, the volume always seems to
increase from a WAV file that is max'ed but no clipping shows in SoundForge.
Once ripped, the Mp3's all seem to clip and increase in volume when checked
in SoundForge 7.
What I've been doing is reducing the volume of the WAV to around -1.03db
(sometimes another -0.9db a few times more) in S.F. and then ripping to Mp3
it using S.F. So far clipping is gone doing it that way but using the other
software mentioned above that depend on LAME 3.96 seem to increase volume
until clipping occurs. Decreasing their volume just brings the flat-tops
(clips) down but now the Mp3 are distorted as well.
It's a pain to manually decrease volume of the WAV and rip each separately
in S.F. Any other better (i.e. faster) way to avoid "Increased volume and
clipping" of ripped files?
The filter banks used in MP3 are, IIRC, phase and amplitude complementary, so
ringing does not really enter into the picture. However, the lossy encoding
process adds a LOT of quantization noise to the individual frequency bands. I
believe that this added noise is what increases the peak levels. (It should
not increase perceived loudness, just peak levels.)
Note that if the psychoacoustic model is working correctly, you won't hear the
added noise because it is added in places where it will be masked by the
original program material. At lower bit rates, this assumption starts to break
down and the noise starts to become unmasked because the noise-to-mask ratio
starts to become positive in certain frequency bands. The lower the bit rate,
the more the peak level will increase, for exactly the same reason--lower bit
rates mean more added quantization noise.
There is no real solution for this problem other than to allow some headroom
before the encode process.
|