Head-Fi.org › Forums › Equipment Forums › Sound Science › High shelf filters and lossy/lossless tests
New Posts  All Forums:Forum Nav:

High shelf filters and lossy/lossless tests

post #1 of 15
Thread Starter 

I've been intrigued by seeing several tests where lossy recordings, often at moderate bitrates, do better than lossless versions in testing, and I wonder if I've stumbled on an explanation:

 

 

 

Quote:

http://people.xiph.org/~xiphmont/demo/neil-young.html

 

 

there is no point to distributing music in 24-bit/192kHz format. Its playback fidelity is slightly inferior to 16/44.1 or 16/48, and it takes up 6 times the space. There are a few real problems with the audio quality and 'experience' of digitally distributed music today. 24/192 solves none of them. While everyone fixates on 24/192 as a magic bullet, we're not going to see any actual improvement...

 

192kHz digital music files offer no benefits. They're not quite neutral either; practical fidelity is slightly worse. The ultrasonics are a liability during playback.

 

Neither audio transducers nor power amplifiers are free of distortion, and distortion tends to increase rapidly at the lowest and highest frequencies. If the same transducer reproduces ultrasonics along with audible content, any nonlinearity will shift some of the ultrasonic content down into the audible range as an uncontrolled spray of intermodulation distortion products covering the entire audible spectrum. Nonlinearity in a power amplifier will produce the same effect. The effect is very slight, but listening tests have confirmed that both effects can be audible.

 

 

My thought is that lossy recordings are removing high frequency information that isn't very important - and may well be over many testers hearing range even if not literally ultrasonic - and which when present causes distortions either in output devices (ie speakers, iems, etc) or even the ear itself. 
 
 
post #2 of 15

The better codecs (AAC, LAME) don't alter the frequency response unless the bitrate is very low.

post #3 of 15
Quote:
Originally Posted by scuttle View Post

I've been intrigued by seeing several tests where lossy recordings, often at moderate bitrates, do better than lossless versions in testing, and I wonder if I've stumbled on an explanation:

 

Neither audio transducers nor power amplifiers are free of distortion, and distortion tends to increase rapidly at the lowest and highest frequencies. If the same transducer reproduces ultrasonics along with audible content, any nonlinearity will shift some of the ultrasonic content down into the audible range as an uncontrolled spray of intermodulation distortion products covering the entire audible spectrum. Nonlinearity in a power amplifier will produce the same effect. The effect is very slight, but listening tests have confirmed that both effects can be audible.

 

My thought is that lossy recordings are removing high frequency information that isn't very important - and may well be over many testers hearing range even if not literally ultrasonic - and which when present causes distortions either in output devices (ie speakers, iems, etc) or even the ear itself. 
 
 

What you've stumbled on isn't an explanation, it's an opinion without substantive evidence.  

 

The statement in bold is very black/white, and makes no allowance for the possibility that power amps can be highly linear, and pass the tiny ultrasonics found in some high-rate files without causing IMD, and that most transducers will just fail to respond to signals outside their range.  

 

The big question is: what listening tests have confirmed that both effects can be audible?  

The smaller question is: if the can be, are they typically, or is it a case of the extremes floating to the top?

 

If they cite the tests, we might stand a chance of evaluating the statement, otherwise its just opinion.  

post #4 of 15

CDs or their lossless versions don't contain much above 20 kHz either. Why should e.g. an mp3 do better if it rolls off a bit earlier?

post #5 of 15
Thread Starter 
Quote:
Originally Posted by jaddie View Post

What you've stumbled on isn't an explanation, it's an opinion without substantive evidence. 

 

 

What I wrote was "I wonder if I have stumbled on an explanation." I.e. this is offered as a hypothesis for testing, not certain fact. The important thing about hypothesis is that it should be reasonably testable, which I think this one is.

 

The statement in bold is very black/white, and makes no allowance for the possibility that power amps can be highly linear, and pass the tiny ultrasonics found in some high-rate files without causing IMD, and that most transducers will just fail to respond to signals outside their range.

 

This is faulty logic: it merely means that if you are correct that there is a subset of amps and transducers that will not effected - which obviously is not enough to stop a statistically measured effect, but only to reduce it.

 

Also I was explicit that I was considering the harmonics from sound above ~15kHz, which is NOT "ultrasonic" and which audio equipment is designed to respond to. I.e.

 

My thought is that lossy recordings are removing high frequency information that isn't very important - and may well be over many testers hearing range even if not literally ultrasonic - and which when present causes distortions either in output devices (ie speakers, iems, etc) or even the ear itself. 


Edited by scuttle - 2/18/13 at 3:45am
post #6 of 15
Thread Starter 
Quote:
Originally Posted by xnor View Post

CDs or their lossless versions don't contain much above 20 kHz either. Why should e.g. an mp3 do better if it rolls off a bit earlier?

 

Ahem:

 

My thought is that lossy recordings are removing high frequency information that isn't very important - and may well be over many testers hearing range even if not literally ultrasonic - and which when present causes distortions either in output devices (ie speakers, iems, etc) or even the ear itself. 

 

Possibly I was taking too much for granted, but the point here is that removing higher frequencies (say over 15kHz) in what is normally considered the audible range (20-20,000Hz) may result in higher sq through lower distortion products. And that this may well be cost free in blind sq tests because a statistically significant number of listeners can't hear the information removed - so they benefit from reduced distortion at no cost.

 

..Come to think of it, isn't this what the fr graphs of the HM601 and 801 show them doing???

post #7 of 15

Oh I see what you mean. 15 kHz seems a bit extreme though. You'd have to have high frequency hearing loss to not notice it in an ABX test.

 

I wouldn't compare a "smooth" roll off to a brickwall filter (sudden cut off).

 

 

Let me take the idea to the extreme: bandlimiting the audio to a single frequency, for example 1000 Hz, would eliminate intermodulation distortion. biggrin.gif

post #8 of 15
Thread Starter 
Quote:
Originally Posted by bigshot View Post

The better codecs (AAC, LAME) don't alter the frequency response unless the bitrate is very low.

 

How low is "very low"? It is somewhat lower br recordings I'm interested in - eg the recent 192 vs 320 test.

 

Also:

 

- I'm not sure the fr is the important thing here. What may matter is the total AMPLITUDE of unheard (by middle aged ears) sound that causes distortion products.

 

- The Page Of LAME is exactly an example of clarity but it seems to say that by default LAME uses a low-pass filter at low br:

 

 

http://lame.sourceforge.net/gpsycho.php

 

Lowpass filtering based on the compression ratio. For high compression ratios, low pass filtering will improve the results. The exact amount of filtering needed depends on the music and personal preferences - the formula to decide how much lowpass filtering to use may need some tuning. At 256kbs, no filterings is done. At 128kbs, the lowpass filter is around 15.5khz.

 

post #9 of 15
Quote:
Originally Posted by jaddie View Post

What you've stumbled on isn't an explanation, it's an opinion without substantive evidence.  

 

The statement in bold is very black/white, and makes no allowance for the possibility that power amps can be highly linear, and pass the tiny ultrasonics found in some high-rate files without causing IMD, and that most transducers will just fail to respond to signals outside their range. 

Sure, a tweeter won't reproduce bass frequencies and a bass woofer won't reproduce treble but feeding the driver frequencies that are clearly out of its comfort zone is a bad idea. Either sound quality will be adversely affected or the driver will be damaged. I think that is fact.

 

Are tweeters in hi-fi speakers made to reproduce frequencies as high as 40 or even 90 kHz?

post #10 of 15

It seems the lower the bitrate, the higher the peak loudness for whatever that's worth.

post #11 of 15
Quote:
Originally Posted by scuttle View Post

 

How low is "very low"? It is somewhat lower br recordings I'm interested in - eg the recent 192 vs 320 test.

 

128 and below I believe. At 192 and above, the response should be stone flat.


Edited by bigshot - 2/18/13 at 11:58am
post #12 of 15

It seems to be hard to get concrete information on this. I've seen this chart posted:

 

kbps lowpass
8, 2000
16, 3700
24, 3900
32, 5500
40, 7000
48, 7500
56, 10000
64, 11000
80, 13500
96, 15300
112, 16000
128, 17500
160, 18000
192, 19500
224, 20000
256, 20500
320, 21000

 

Here by a xclame dev.

 

Not to mention that LAME would remove a bunch of really high frequency information even without an explicit lowpass filter, since it's not very audible(or completely inaudible) in most cases.


Edited by chewy4 - 2/18/13 at 12:14pm
post #13 of 15

That chart pretty much matches what I hear. With AAC, it might be a bit better.

post #14 of 15
Quote:
Originally Posted by chewy4 View Post

It seems to be hard to get concrete information on this. I've seen this chart posted:

 

kbps lowpass
8, 2000
16, 3700
24, 3900
32, 5500
40, 7000
48, 7500
56, 10000
64, 11000
80, 13500
96, 15300
112, 16000
128, 17500
160, 18000
192, 19500
224, 20000
256, 20500
320, 21000

 

Here by a xclame dev.

 

Not to mention that LAME would remove a bunch of really high frequency information even without an explicit lowpass filter, since it's not very audible(or completely inaudible) in most cases.

Just tried a few tests to confirm the above chart.  Assuming those are rates for stereo files, I find the LPF filter to be considerably higher than listed for low bit rate files using the mp3 encoder in iTunes, with spectrum analysis in Audacity.  For example, the 16K rate is shown as 3700, my test file made it up to 4500 before the roll-off starts.  Not a big deal, really, but when looking at bit rates that result in 20KHz response it looks like you could us lower rates than shown before loosing anything below 20KHz.  Probably depends on the specific codec, no idea which one is in iTunes, but I thought it was interesting. Also, it looks like at low rates mp3 has a higher top end than AAC, at high rates it doesn't seem to make any difference.

post #15 of 15

RIght after I posted the above I realized this could be entirely program dependent, because of how perceptual coding works. I might have just found a bit of music with highs that won't mask well, so they get coded instead of removed.  Looks like we'd need a group of files, and an average.  Sorry. 

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › High shelf filters and lossy/lossless tests