Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Audibility of "typical" Digital Filters in a Hi-Fi Playback  (Read 367694 times) previous topic - next topic
0 Members and 24 Guests are viewing this topic.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #275
In my last post I was guilty of using the royal we, though I can think of a few members who'd see at least a little irony in what I quoted.

Since I'm on the subject of we...

I haven't seen this awareness in other forums.  Is there a link here where folks have said that in double blind tests, people can tell the difference between how 24 bit files are or are not dithered to 16 bits?

I don't understand this tendency to use  third party ghosts in such references.  Who are you speaking on behalf?    Let's have that list of people that agree with what you said.

More disappointing personal commentary.  Looks like HA forum is much less professional in this manner than I thought prior to joining the forum

It is not our responsibility to acquaint you with our community.  We have a search function.  No one is prohibiting you from using it prior to posting.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #276
Guys, there is a difference between linear spectral density (measured in V/sqrt(Hz)) and linear spectrum (measured in V).
The density is obviously lower for a wider bandwidth, since it is "spread" across the wider bandwidth.

sqrt(22050)/sqrt(96000) = -6.38 dB
"I hear it when I see it."

 

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #277
For months on AVS Forum we have been discussing the audibility of high resolution audio.  After passing ABX DBT after ABX DBT, the claim always went back, "well, Meyer and Moran didn't find this difference and it was peer reviewed so this data must be wrong."


I guess you didn't get the memo Amir (silliy I should expect you to comprehend it, given that it is in a recent post to this thread), but the ABX tests I posted on AVS had a built in flaw that I only very recently became aware of which is that the transition band of my downsampling filter was way too narrow per this certain peer reviewed paper from Meridian.

It was covered in the second part of this post: http://www.hydrogenaud.io/forums/index.php...st&p=881145

We know that the other set of files had other serious flaws of their own.

It thus becomes improper for anybody to say: "After passing ABX DBT after ABX DBT" because there were no proper tests.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #278
the ABX tests I posted on AVS had a built in flaw

Any reasonable, objective person would see a flawed test as flawed and thus the results flawed, be they online games or manufacturer doctored up results.
Such a position is incompatible with objectivity, if one accepts flawed, unsupervised-gameable and deliberately doctored results like this BS test.
It must be something else.

cheers,

AJ
Loudspeaker manufacturer

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #279
For months on AVS Forum we have been discussing the audibility of high resolution audio.  After passing ABX DBT after ABX DBT, the claim always went back, "well, Meyer and Moran didn't find this difference and it was peer reviewed so this data must be wrong."


I guess you didn't get the memo Amir (silliy I should expect you to comprehend it, given that it is in a recent post to this thread), but the ABX tests I posted on AVS had a built in flaw that I only very recently became aware of which is that the transition band of my downsampling filter was way too narrow per this certain peer reviewed paper from Meridian.

It was covered in the second part of this post: http://www.hydrogenaud.io/forums/index.php...st&p=881145

We know that the other set of files had other serious flaws of their own.

It thus becomes improper for anybody to say: "After passing ABX DBT after ABX DBT" because there were no proper tests.

I appreciate you falling on your sword this way as to invalidate your own test that you had put out there for years to prove high resolution audio is not differentiated against CD rate.  But it is not necessary as your theory is not correct here.  As you know, the fidelity of your resampler was questioned repeatedly and you defended it.  This is one such example with me asking the same question: http://www.avsforum.com/forum/91-audio-the...ml#post26134194

Quote
Why are you asking me? Won't I be a biased source? ;-)

If you read my posts you know that the SRC I use was the one in Cool Edit Pro 2.1 with pre/post filtering, and you know about the SRC reviews at Infinite Wave. Adobe Audition 2.0 is a close descendant of Cool Edit Pro 2.0 which is testsed at Infinnite Wave. I could tell you to go figure.

However, I can also tell you that I've run the tests that Infinite Wave used on all those SRCs on the Cool Edit Pro 2.1 SRC with very similar results. I also did some tests that are far more difficult than what Infinite Wave used, and found it has a weakness. The weakness is that if you upsample or downsample with really high ratios such as upsampling from 44.1 to several, maybe 10 megahertz, its digital filter's bandpass filtering seems to loose some of its precision. The circumvention is to do the upsampling in steps.


Strange that such obvious problems as to invalidate your test were not brought out in the above testing.  Perhaps your testing wasn't comprehensive enough then?

You say that the fault lay with using the quality slider in advanced settings and pushing it to max.  As it turns out, I had tested it on the default 50% setting as I reported here: http://www.whatsbestforum.com/showthread.p...ll=1#post281829

Quote
Arny's files never had sync problems.  But yes, I did resample them using latest version of Audition CC and I could still tell them apart. 

============

foo_abx 1.3.4 report
foobar2000 v1.3.2
2014/07/24 20:27:41

File A: C:\Users\Amir\Music\Arnys Filter Test\keys jangling amir-converted 4416 2496.wav
File B: C:\Users\Amir\Music\Arnys Filter Test\keys jangling full band 2496.wav

20:27:41 : Test started.
20:28:07 : 00/01  100.0%
20:28:25 : 00/02  100.0%
20:28:55 : 01/03  87.5%
20:29:02 : 02/04  68.8%
20:29:12 : 03/05  50.0%
20:29:20 : 04/06  34.4%
20:29:27 : 05/07  22.7%
20:29:36 : 06/08  14.5%
20:29:44 : 07/09  9.0%
20:29:55 : 08/10  5.5%
20:30:00 : 09/11  3.3%
20:30:07 : 10/12  1.9%
20:30:16 : 11/13  1.1%
20:30:22 : 12/14  0.6%
20:30:29 : 13/15  0.4%
20:30:36 : 14/16  0.2%
20:30:41 : 15/17  0.1%
20:30:53 : 16/18  0.1%
20:31:03 : 17/19  0.0%
20:31:07 : Test finished.

----------
Total: 17/19 (0.0%)


So I managed to differentiate the files with a much newer implementation of your SRC, at less than maximum setting and still managed to easily tell the files apart in DBT ABX test.

------
But let's put all of this aside and presume you are right.  This solidly makes the case for getting the original bits prior to down conversion to 16/44.1.  After all, if you with all of your experience would produce an obviously faulty conversion as to have clear audible distortions, and said problem was setting a "quality slider to 100%," what hope is there for any content producer creating transparent content?  Are we to assume they know far more than you do as to not have made this mistake Arny?

Let me leave you with this note from Adobe Audition CC's Help file: http://help.adobe.com/en_US/audition/cs/us...0c2c5-7f52.html

In the Advanced section, drag the Quality slider to adjust the quality of the sampling conversion.
Higher values retain more high frequencies, but the conversion takes longer. Lower values require less processing time but reduce high frequencies.

Use higher Quality values whenever you downsample a high rate to a low rate. When upsampling, higher values have little effect.


This puts it at odds with your post in the above link:

BTW, there is a lesson here. The Quality slider on software resamplers such as CoolEdit Pro/Audition is a tuning knob for transition band width. Higher quality means a narrower transition band.

If you push it to the far right (99%) the resulting transition band became  exceedingly narrow in my tests - just a few Hz, which puts it in the same realm as the signal that Meridian used for listener training.

If you push it to the far left the transition band is more like 1.6 KHz and better approximates a real world converter.

All of the sample rate ABX  test files I've circulated lately were made with quality set for the maximum. They are therefore not representative of real world circumstances. One word: Invalid.


But what do I know. 
Amir
Retired Technology Insider
Founder, AudioScienceReview.com

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #280
... But what do I know. 


That question appears to have been answered quite comprehensively, at least in the subjects covered by this thread.
Regards,
   Don Hills
"People hear what they see." - Doris Day

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #281
But what do I know. 

Hopefully that unsupervised gameable online computer tests (including those by ex-MS execs) are not the subject of the thread and don't prove anything related to 16/44 transparency.
Speaking of which, Amir how do you explain this:


Amir, which Meridian/BS statement is false, the one above from their website about "achieving 16/44 transparency" , or the new BS test just released claiming (apparent) 16/44 non-transparency (as we both know due to RPDF dither doctoring or possible setup artifacts)?
They can't both be right.

cheers,

AJ
Loudspeaker manufacturer

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #282
It is not our responsibility to acquaint you with our community.  We have a search function.  No one is prohibiting you from using it prior to posting.

Ah, I didn't realize that.  Thanks for the suggestion.  So I searched for noise modulation audibility and landed on this thread: http://www.hydrogenaud.io/forums/index.php?showtopic=16963.  Yet another set of excellent posts by David.  You guys have a gift there.  He has super hands on experience with signal processing.  I hope he doesn't mind me reaching out to him as I often run into start-ups I am advising that could benefit from his expertise and my Microsoft resources are too busy to help out.

Anyway, here is the key post by another member I sadly don't know but seems to also be knowledgeable:

If I'm not wrong (2Bdecided probably knows better than me), mathematically, with rectangular dither, 1 bit amplitude is enough to totally remove distortion, but noise modulation remains. With triangular dither, 1 bit is enough to totally remove both, distortion and noise modulation. With noiseshaping dither, the amount depends on the shaping of the noise floor, but is usually below 1 bit. I don't know of a way of calculating this, but maybe the mentioned paper does. At the tests I did on the mentioned thread I used subjective listening tests for the noise modulation issue, using a critical test signal, amplified several tenths of dBs to make dither failure audible, and also frequency analysis (FFT) to detect distortion.


So when you said most everyone here agrees noise modulation is audible, you meant it under these conditions?  FFT analysis, and amplifying the signal by several tens of dBs?  Is it your impression that Stuart's test include such overamplification and FFT analysis was performed by expert listeners who understand what to look for?

AJ, is this the audible problem you think should have been avoided in the test?
Amir
Retired Technology Insider
Founder, AudioScienceReview.com

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #283
So when you said most everyone here agrees noise modulation is audible

Since I'm the only other person you quoted, I'd like to see where I said this.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #284
AJ, is this the audible problem you think should have been avoided in the test?

I think you should quit evading/dancing around the doctored up dither BS test dilemma and obvious conflict with BS statements direct from Meridian above. Both can't be right.
Plus the whole artifacts issue with unconfirmed system switching software/speaker transparency, just like BS and JJ warn about.
But I know you.  ..and now so does HA.

cheers,

AJ
Loudspeaker manufacturer

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #285
So when you said most everyone here agrees noise modulation is audible

Since I'm the only other person you quoted, I'd like to see where I said this.

I was going to say you could do your own search but thought it would not be in good form .  So here it is: http://www.hydrogenaud.io/forums/index.php...mp;#entry880450


all this paper seems to show is that if you don't dither at all, or dither improperly, a group of carefully trained listeners will, on some sample material, be able to distinguish the 16-bit content from the original content

Right, and most of us were already aware this was possible.[/color]


Still stand by that position given how it is at odds with the post I linked to on this forum?


Amir
Retired Technology Insider
Founder, AudioScienceReview.com

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #286
About the audibility of improper dither, yes.

About all that the paper seems to show, no; though this has been made clear already.

Perhaps you should read it more carefully noting the qualifications on the statement (or lack thereof).

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #287
About the audibility of improper dither, yes.

Yes what?  That it takes tens of dbs of level amplification and FFT examination to tell?
Amir
Retired Technology Insider
Founder, AudioScienceReview.com


Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #289
About all that the paper seems to show, no; though this has been made clear already.

Nothing remotely has been accomplished there let alone be clear.


Please explain how that green line is jumping up to exceed the threshold of hearing. 

Please explain likelihood of peer review not realizing what you have made clear in this thread.

Please explain how these the manifestations of the problem you see as reported in the paper by the subjects taking a blind test:

It was reported that filtering gave "softer edges" to
the instruments, and "softer leading edges" to musi-
cal features with abrupt onsets or changes. Echoes,
when audible, were identi ed as being affected the
most clearly by the ltering. It was felt that some
of the louder passages of the recording were less ag-
gressive after filtering, and that the inner voices (sec-
ond violin and viola) had "a nasal quality". Over-
all, the filtered recording gave a "smaller and
flatter auditory image", and speci cally the physical space
around the quartet seemed smaller.

Listeners described that quantization gave a "rough-
ness" or "edginess" to the tone of the instruments,
and that quantization had a significant impact on
decay, particularly after homophonic chords, where
"decay was sustained louder for longer and then died
suddenly". This could be an effect of quantization
distortion; it is interesting that this was audible even
in a 24-bit system, and is consistent with the hy-
potheses of Stuart [29] that 16 bits are not sucient
for inaudible quantization.


Quote
Perhaps you should read it more carefully noting the qualifications on the statement (or lack thereof).

Perhaps.  But you need to be specific and quote the section of the paper.  And indicate that you are in possession of the paper and understand its content.
Amir
Retired Technology Insider
Founder, AudioScienceReview.com


Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #291
FFS, we each don't communicate in a vacuum. You've already been schooled by others on the matter.

Any test scores showing rectangular dither didn't affect the results?

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #292
To summarize: the recent Meridian tests sabotaged 16/44 and 16/48 processing by intentionally using

(1) suboptimal dither PDF,
(2) suboptimal dither PSD,
(3) and suboptimal low pass filter transition band width.

All fully documented in this thread.
I agree that all of these things are sub-optimal, but I think we* should at least have the humility to accept that we* wouldn't normally expect any of them to be audible.

(we* = people who read+post at HA a lot.)

For each separate issue, it is of course possible that the exact processing parameters and the exact playback parameters have combined to bring these issues up to the threshold of audibility. I'm not yet clear whether we have sufficient information to make this judgement, but I think that's the judgement we should attempt to make.


I'm especially interested to see you criticising the narrow filter transition band Arny. If the width of the filter transition band matters when that filter is acting at 24kHz, then that raises all kinds of interesting questions. I know why dither matters if it's at an audible level: you can hear it. I don't know why filter ringing matters when it's at an ultrasonic frequency. I know various possible mechanisms by which it might become audible, but that's different from saying that it is.

Cheers,
David.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #293
We talk about filters and PCM here.
Keep in mind the modern audiophile has no doubt DSD sounds better all the time as PCM because its differernt filter behaviour.



Interestingly enough, the AES Conference paper that centerpieces this thread pretty well trashes DSD:

"
Another potential problem with the setup used by
Meyer and Moran [4] concerns the source material.
SACD uses 1-bit DSD (Direct-Stream Digital) en-
coding, which supports a wide dynamic range for
frequencies below about 20 kHz by \noise shaping",
that is, pushing the noise in the signal to frequencies
higher than 20 kHz. As a result, the signal-to-noise
ratio for frequencies higher than 20 kHz decreases
quickly. Therefore SACD is not a suitable candi-
date for the kind of listening test intended to estab-
lish any contribution to perception from frequencies
higher than those encoded by a conventional CD. In
addition, it is customarily recommended that SACD
players include a high-order low-pass lter in the re-
gion of 30-50 kHz to reduce the high-frequency noise
introduced to ampli fiers and tweeters; such fil lters
may have audible impact.
"

Talk about falling on your sword!

Of course Meridian profits from so-called high resolution PCM with their data compression patents (MLP), not DSD.

So it is not their sword that they are falling on, it is Sony et al that takes the fall.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #294
To summarize: the recent Meridian tests sabotaged 16/44 and 16/48 processing by intentionally using

(1) suboptimal dither PDF,
(2) suboptimal dither PSD,
(3) and suboptimal low pass filter transition band width.

All fully documented in this thread.
I agree that all of these things are sub-optimal, but I think we* should at least have the humility to accept that we* wouldn't normally expect any of them to be audible.

(we* = people who read+post at HA a lot.)

For each separate issue, it is of course possible that the exact processing parameters and the exact playback parameters have combined to bring these issues up to the threshold of audibility. I'm not yet clear whether we have sufficient information to make this judgement, but I think that's the judgement we should attempt to make.


I'm especially interested to see you criticizing the narrow filter transition band Arny.



Don't get me wrong. I don't know what to think. I'm just reading "Peer reviewed" AES papers...

There are many assertions in the recent Meridian paper that IMO need far better support. This is just one of them.

The main point is that if Meridian critiizes flat PSD dither and non-TPDF dither when they are selling digital processors,  why are they ramming them down our throats when they are testing other people's products?

If they use files created by low pass filters with too-narrow transition bands to train their listeners to hear filtering artifacts, why are they ramming them down our throats when they are testing other people's products?

Looks like meat for some ABXing.

Quote
If the width of the filter transition band matters when that filter is acting at 24kHz, then that raises all kinds of interesting questions. I know why dither matters if it's at an audible level: you can hear it. I don't know why filter ringing matters when it's at an ultrasonic frequency. I know various possible mechanisms by which it might become audible, but that's different from saying that it is.


That all goes through my mind every time I try to think about their apparently unsupported claims about transition bands.

But it must be true - the AES review board signed off on it and gave the paper their highest reward, right? ;-)

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #295
Here's what is known about the recording

"The recordings used were selections from The signals
used were extracts from Haydn's String
Quartet Op.76 No.5 in D \Finale, Presto" from
\Nordic Sound (2L Sampler)" 1, issued with a sample
rate of 192 kHz using 24-bit PCM23.

This recording can be downloaded in a number of different formats from here:

https://shop.klicktrack.com/2l/35847

Price in 24/192 format: $6.00
No one has mentioned something really funny.

amirm wants the original bits.

This is reportedly a 352.8kHz/24bit/5.1 channel recording.

Check out the available formats. You can't buy the original 352.8kHz/24bit/5.1. The "best" available options are downgraded to stereo, "downgraded" to 96kHz, and "downgraded" to DSD. You can't buy the original bits. At most, you get something with 1/3rd the number of bits, or less than half the number of channels.

The BluRay (which is cheaper) includes 192kHz/24bit/5.1.


Is anyone going to post a 30 second sample?

Cheers,
David.

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #296
Quote
Now I don't think that the choice of dither usually matters, but in critical evaluations of the limitations of a format I would at least expect triangular dither.

If it usually doesn't matter then it is not an "obvious flaw."

Let me rephrase so even you can understand it:
In an evaluation of the limitations of a format, which normal listening is not, you shouldn't go for anything subpar, be it dither, filters, playback hard/software ...

Now I don't know if dither was a problem. But it is reasonable to have doubts when these people have written papers on dither and noise shaping and how it can be audible with 16 bits!


Yes, another win for science, and a fail for laymen trying to interpret the same.  The test is run at 192 Khz Arny.

And since you have the algorithms at hand, you know exactly how they did or didn't account for noise bandwidth in their 44.1/16 simulation within a 192 kHz format, right? Right?
No..

Also, I've already explained it, but do you know the difference between spectral density and an amplitude/linear spectrum?
You can plot the spectral density of rectangular dither with 192/16 at -144 dB (which is what they did in that fig. 3), but the SNR is still only ~95 dB.
"I hear it when I see it."

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #297
So if you plot the dither noise in fig. 3 in terms of human-audible significance (approx. 1/3rd octave wide noise-gathering bandpass filters above 1 kHz), you will see that around ~4 kHz we are down roughly -100 dB relative to the mean acoustic gain, so above even the hearing threshold for single tones.
And that doesn't account for additional attention-drawing due to noise modulation.
"I hear it when I see it."

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #298
Please explain how these the manifestations of the problem you see as reported in the paper by the subjects taking a blind test:

[...]

Listeners described that quantization gave a "rough-
ness" or "edginess" to the tone of the instruments,
and that quantization had a significant impact on
decay, particularly after homophonic chords, where
"decay was sustained louder for longer and then died
suddenly". This could be an effect of quantization
distortion; it is interesting that this was audible even
in a 24-bit system, and is consistent with the hy-
potheses of Stuart [29] that 16 bits are not suffcient
for inaudible quantization.


First of all, the above is subjective and anecdotal. But anyway, some of it doesn't sound too far fetched. I guess you do not know what noise modulation even is.

Secondly, Stuart has written papers that show that 52 kHz / 11 bits are the absolute minimum PCM channel using noise shaping, capable of replicating the information received by the ear. 14 bits "ought to be adequate" to offer enough headroom. That would be 18.2 bits for a rectangular channel (TPDF, no noise shaping).
If proper shaping can make a few bits difference, then why do you dismiss the idea that RPDF could be inadequate in such a test given the 16-bit quantization?

"I hear it when I see it."

Audibility of "typical" Digital Filters in a Hi-Fi Playback

Reply #299
Here's what is known about the recording

"The recordings used were selections from The signals
used were extracts from Haydn's String
Quartet Op.76 No.5 in D \Finale, Presto" from
\Nordic Sound (2L Sampler)" 1, issued with a sample
rate of 192 kHz using 24-bit PCM23.

This recording can be downloaded in a number of different formats from here:

https://shop.klicktrack.com/2l/35847

Price in 24/192 format: $6.00
No one has mentioned something really funny.

amirm wants the original bits.

This is reportedly a 352.8kHz/24bit/5.1 channel recording.

Check out the available formats. You can't buy the original 352.8kHz/24bit/5.1. The "best" available options are downgraded to stereo, "downgraded" to 96kHz, and "downgraded" to DSD. You can't buy the original bits. At most, you get something with 1/3rd the number of bits, or less than half the number of channels.

The BluRay (which is cheaper) includes 192kHz/24bit/5.1.


Is anyone going to post a 30 second sample?
.


Is http://www.hydrogenaud.io/forums/index.php...st&p=881260  a good place?