Skip to main content

Topic: Lowpasses, clipping, and normalization (Read 4774 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • mp3fan
  • [*][*]
Lowpasses, clipping, and normalization
Dibrom,

I recently tested my hearing and found that my frequency range is about 17.3 Khz and then a serious dropoff after that.  So I'm encoding my mp3s with a 17.5 lowpass which should start it's lowpass at 17.2 KHz which is right at my threshold.  I figure this will cut down on wasted bytes.  However I'm now experiencing more clipping which is obviously attributed to the lower lowpass. 

I've found that I need to normalize at 87% to get rid of the clipping on EAC.  I used the MAD audio decoder in winamp to confirm the prescence of clipping was there and eventually eliminated.  Perhaps this information may help you set a proper setting for your lower bitrate commandlines. 

Should I not use normalization?  I'm currently setting the normalization to 87% if the maximum sample in the wav file exceeds 87% to avoid clipping completely.  That's why I'm not using the scale function in LAME to simulate a normalization.  This is because the scale function will treat every file with the scale even if it doesn't exceed a level which could cause clipping.  Hence a wav file with a peak sample of 70% would lose 13% of it's volume unnecessarily.

Thanks,
mp3

  • Dibrom
  • [*][*][*][*][*]
  • Administrator
Lowpasses, clipping, and normalization
Reply #1
I moved this post because it wasn't really related to the other thread it was in.

Quote
However I'm now experiencing more clipping which is obviously attributed to the lower lowpass.


Are you sure the clipping is coming from the lowpass?  Have you tried encoding the exact same file with different lowpasses and actually confirmed that there are more clipped samples in one vs the other?

Quote
Perhaps this information may help you set a proper setting for your lower bitrate commandlines.


I've tuned the lower bitrate presets to prevent clipping via listening tests.  I'm not trying to eradicate all clipping, I'm to eliminate any audible clipping, which I have done for the most part.  To remove all clipping you'd have to use too high high (or rather low) of a scale value.

Quote
Should I not use normalization? I'm currently setting the normalization to 87% if the maximum sample in the wav file exceeds 87% to avoid clipping completely. That's why I'm not using the scale function in LAME to simulate a normalization. This is because the scale function will treat every file with the scale even if it doesn't exceed a level which could cause clipping. Hence a wav file with a peak sample of 70% would lose 13% of it's volume unnecessarily.


If you want to get rid of clipping, use mp3gain.. its lossless and it will handle this thing better.

  • cbuchner1
  • [*]
Lowpasses, clipping, and normalization
Reply #2
Quote
Originally posted by Dibrom

If you want to get rid of clipping, use mp3gain.. its lossless and it will handle this thing better. 


just a quick question, which may have been answered previously (but I missed it).

I assume mp3gain modifies the scale factors in each subband to achieve the scaling.

How can mp3gain be lossless when there is no scalefactor for the last (final) subband? It would have to modify the quantized values - and this "requantization" will always add some additional noise. Is that correct?

  • Dibrom
  • [*][*][*][*][*]
  • Administrator
Lowpasses, clipping, and normalization
Reply #3
I don't quite know the details about how the actual modification itself is performed, but I don't think it has anything at all to do with scalefactors.  There is simply a gain value somewhere that can be modified.

  • mp3fan
  • [*][*]
Lowpasses, clipping, and normalization
Reply #4
MP3 Gain?  How do I use it?  Do I have to download anything?  Confused here.

mp3

  • Dibrom
  • [*][*][*][*][*]
  • Administrator
Lowpasses, clipping, and normalization
Reply #5
There's a link to the program in the mp3 utilities links section on this page.  The relevant information needed is on the homepage.  Mp3gain basically uses some of the Replaygain stuff to adjust gain and remove clipping, etc..

Lowpasses, clipping, and normalization
Reply #6
maybe you shouldnt cut the frequencies above 17,2 khz. but although it is well known we cant hear frequencies higher than some 18 (children up to 20) khz, why is it that a lot audiofiles swear by SuperAudioCD, which goes up to 40 khz??? Its useless? No, its not... the higher freqs have some sort of "wave effect" on the lower, which you actually CAN hear. btw, if a speaker is attached to your head, you can hear freqs upto some 30 khz or something like that, I heard a long time ago. who knows 'bout this? anyway, it something bout this is true, maybe you hear UNCONSIOUSLY these freq's - or whatever...  :confused:

J
[span style=\'font-family:Arial\'][span style=\'color:red\']Life Sucks Deeply[/span][/span]

  • Dibrom
  • [*][*][*][*][*]
  • Administrator
Lowpasses, clipping, and normalization
Reply #7
While I've seen many theories that frequencies above 22khz are somehow "audible" or have an "effect" on the sound, I don't think I've ever seen any definitive proof, as in hard blind listening test data from some sort of scientific establishment, that this is actually the case.  Though it may be out there.  But to be honest, as I've said elsewhere, if for some reason you are a person that can hear with clarity up to 22khz (this is extraordinarily rare), then you shouldn't be using mp3 in the first place.  Right over 16khz you are going to be losing loads of data and the quality significantly degrades.  Considering that, a lower lowpass is going to be the least of your concerns.