 
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
|
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
209.97.232.27
In Reply to: RE: Drums, solo posted by Werner on October 13, 2009 at 11:47:57
Werner, you are awesome!
Those are great clips and very instructive. All of the low-bit versions sound like the original clip but with some other "thing" added:
a) No dither -- sounded like the original drums with severe vinyl mistracking (such as when there is a giant blob of dust on the stylus) added.
b) Dither -- sounded like the original drums with an obnoxious level of hiss added. The hiss was extremely distracting.
c) Noise shaped dither -- same as (b) except that the hiss was nearly inaudible. Amazingly close to the original MP3 file!
Just to satisfy my curiosity, was the bit reduction the equivalent of a truncation or a rounding?
Thanks again for providing these extremely instructive demonstrations!
Follow Ups:
Truncation. This was done in DAW software that doesn't allow full
control of what's happening down below.
For rounding I'd have to run it through SciLab or Matlab.
I don't agree with your assessment of the truncated version. Haven't heard vinyl mistracking in a long time, but that doesn't sound like it. Truncation sounds grotesque,like someone tweaking an AM receiver during a solar storm and with chainsaw trio thrown in.
This said, glad to have been of some use.
bring bac k dynamic range
...try making a demo comparing truncation versus rounding. I think the difference will be interesting.
As far as the vinyl mistracking, I didn't mean when there is distortion on a really loud passage. I meant when you played a really dusty record and towards the end of the side a huge blob of dust would collect on the stylus and it wouldn't make good contact with the groove.
"...try making a demo comparing truncation versus rounding. I think the difference will be interesting."
I have a hunch such a demo would disappoint you.
I also have a hunch why you prefer simple rounding in your particular case, but to confirm this idea I would like to have the coefficients of your
filter. No, not all of them, just a section long enough to be interesting. It's 16 x oversampling, not? Im that case I would be happy with 18-20 subsequent coefficients. You can even tweak the values themselves if you don't want to reveal anything. I'm only interested in the zeroes, actually.
bring bac k dynamic range
I think you might have some problems with your hearing Charles.. The last clip do have reduced hiss, but it is VERY VERY VERY CLEARLY audible. This is through my the speakers in my laptop.
And get off my lawn, dammit!
Hehe. Well, the noise isn't "almost inaudible", but I realise you might not have meant it literally. I'm also not that young anymore (unfortunately).
I just played back the clip while typing at my keyboard. At this position I am about 80 degrees off axis from my speakers. The noise was at 19 kHz. The off-axis response is probably about 30 or 40 dB down at that frequency and angle.
The sample is mp3 at 128kbs. So that means a lowpass filter at just over 15khz. As you can see for yourself, there are not any sounds above this. http://img23.imageshack.us/img23/365/clipaa.jpg
I also want to quote Werner, because I am not sure what he is saying exactly.
"...Something else entirely. The noise shaping used was extremely aggressive, packing most of the noise in a narrow peak around 19kHz (heck, it is almost like tape bias!), which allows the noise level in the band below to drop dramatically relative to the dither case.
Now I was wondering if such a complex noise shape would survive through MP3. The sample was encoded at 128kb/s, which normally filters anything above 15kHz. This happened here too: the MP3 coding removed the 19kHz noise spike, and yet the sound did not suffer. Which is actually also predictable, as the MP3's low pass filter takes the 19kHz spike and averages it into the lower band."
I'm not quite sure what Werner means by "the sound did not suffer?" If there is supposed to be some kind of peak around 19khz, and the mp3 averages that peak into the lower band, well isn't that exactly the definition of suffering? Hence Charles, you are supposed to hear clearly this even if you listened to your speakers backwards. Not looking to argue, just curious as there is something going on here..
I think Charles was speaking comparatively. Of course the noiseshaped version has hiss that is clearly audible. But it is high-pitched and it is still at a much much lower level than the wideband roar of the flat-dither version. Which is all exactly as the theory predicts, of course.
--
Something else entirely. The noise shaping used was extremely aggressive, packing most of the noise in a narrow peak around 19kHz (heck, it is almost like tape bias!), which allows the noise level in the band below to drop dramatically relative to the dither case.
Now I was wondering if such a complex noise shape would survive through MP3. The sample was encoded at 128kb/s, which normally filters anything above 15kHz. This happened here too: the MP3 coding removed the 19kHz noise spike, and yet the sound did not suffer. Which is actually also predictable, as the MP3's low pass filter takes the 19kHz spike and averages it into the lower band.
Just as our ears do anyway.
bring bac k dynamic range
Post a Followup:
| FAQ |
Post a Message! |
Forgot Password? |
|
||||||||||||||
|
||||||||||||||
This post is made possible by the generous support of people like you and our sponsors:
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2025, Audio Asylum, all rights reserved ]
