No, there were no high scores at all, as Krabapple said above :



Thus retesting was not needed. Which answers 2tec original question : both average results and individual results were taken into account. There was no positive result.

In the Detmold university listening test, 200 listeners took the same challenge. Some scored above the significance threshold, but this was coherent with random guessing at the collective level.
However, one of them got a score of 20/20, which is significant even in a collective test with 200 listeners.
The authors said that unfortunately, a small noise at the beginning of one of the samples, though unheard by the listeners, may have biased the result.
Maybe also this listener really hears ultrasounds... I don't remember the study talking about his or her hearing ability in high frequencies.

Guru ABXed artifacts of mp3, especially on low volume classical and natural instrument music. I think I have followed all of his ABX threads and I don't recall anything about him ABXing frequencies. Ability to ABX an mp3 vs CD doesn't mean it is the frequency that you can distinguish. In fact with all (pre)echo and such artifacts with mp3, higher frequencies are probably the least noticeable/contributing factor.

All resampling is not equal. Compare the Adobe Audition Sweep tone resampling (using proper pre/post filters, or even without the filters) against that of Gold Wave for an easy explanation. Audacity (High-quality Sinc Interpolation) is significantly worse.
http://src.infinitewave.ca/

The resampled triangle sample from Audacity isn’t quite so colorful as the resampled sweep tone, but when comparing Audacity’s result to CoolEdit resampling (the precursor to Audition, for those who don’t know), the visual differences, especially in the critical midbands, are very obvious.

In situations like this I think it's worth mentioning (again) that your soundcard may be resampling during these listening tests as well.

The resampling quality in Audacity, when I tested it, was rather poor. I initially thought it was using libsamplerate (secret rabbit code), but apparently the use another library, based on the same algorithm, due to some licensing issues. Libsamplerate 0.1.3 works very well, and should give excellent results, and you can build audacity on Linux to link against it.

I mentioned the visual differences due to Audacity's poor resampling, since not everyone has decent analysis software. The auditory differences in the triangel sample are so striking that anyone should be able to hear them, no special software required.

Playing with Audacity a little, I found something rather strange, or maybe not too strange as I haven’t had reason to investigate other programs since CoolEdit does such an excellent job.

Originally I generated a sweep tone in CoolEdit at 96kHz (100Hz to 48kHz Sine wave over 10 seconds). I just modified some settings I’d used some time ago. The Sine wave was modulated. I don’t remember the Modulated By value but the Modulation Frequency was 10Hz. Resampled in Audacity, this produced a very colorful Spectral View showing much harmonic distortion, aliasing, and spurious frequencies.

Afterwards I generated the same sweep tone without any modulation, as pure a sine wave as CoolEdit can generate. The Audacity resampling Spectral View of that looks almost like the CoolEdit resampling, very different than the modulated sweep tone. While neither are music, I would say the modulated tone is more representative of most music.

Two different performances would definitely invalidate the test, and the only other means to compare the same recording without introducing heinous variables, is to either dither the 24 to 16, or record a live performance with two A/D converters, one set to 16bit and one to 24, from the same microphone input, at the same sample rate.

Regarding 16 vs 24 bits. I remember reading that our hearing distinguishes the differences at 18 bits resolution but no more, so 16 bits is a bit too low but 24 bits is simply an overkill. With a proper dithering, though, 16 bits is 'about as good as' 18 bits. I guess that does mean that Red Book format is sufficiently close to our hearing threshold in terms of resolution. In terms of sampling rate, 44.1Khz providing 22 Khz of bandwidth is more than enough.

You know what, I'll do this tonight or tomorrow... rip a 24-bit/48kHz DVD track (it'll be Sonic Youth, because its the only LPCM concert DVD I have) and then provide samples at:

24/48
16/48
24/44.1
16/44.1

I'll use SSRC and dither w/noise shaping for the 16-bit versions.

This high resolution track is free : http://www.hydrogenaudio.org/forums/index....showtopic=35624

Working link through Megaupload, post 31, page 2.

I had a quick listen to a 44.1KHz version (which I created with Audition 3.0) but there was no obvious difference for my ears, compared with the original 96KHz version. (I have never found the sound from guitar strings easy to detect deficencies in. I am not saying there are no differences in this laid back performance between the original 96KHz version and a 44.1KHz conversion; but merely that nothing 'stuck out' when I listened to the two versions.)

When I returned to the 96Khz/24bit sample of a triangle being struck, which I commented on at post #50 above, the differences were quite stark, so that is the sound sample I have selected for closer 'amateur analysis' ...



Thanks greynol, cabbagerat and AndyH for your comments.

I have now resampled the 96/24 struck triangle sound using Audition 3.0, set to maximum conversion quality (999). This has not prevented the resampled sound sounding different. I obtained the following results:

A. Original 96KHz resampled to 44.1KHz:- duller, and sound appears to come more from from the left.
B. Original 96KHz resampled to 48KHz:- duller, and sound appears to come more from the left.
C. Above version A resampled back to 96KHz:- no improvement.
D. Above version B resampled back to 96KHz:- no improvement.
E. Original 96KHz resampled to 192KHz:- the 192KHz version had a slightly brighter sound.

The differences between A, B, C or D and the original 96/24 sample were quite stark, and I did not perform ABX tests. Either factor -- change in position of the stereo image, or the loss of apparent high frequencies -- was quite noticeable.

The difference in situation E was relatively slight, so I ABXd, in order to satisfy forum guidelines. I was a little surprised to hear a difference in situation E. I had thought the filter performance in the audible range (and even a bit beyond that) would have been indistinguishable as between a sampling rate of 96KHz and a sampling rate of 192KHz.

Listening devices used
The differences could be heard using the analogue outputs of the motherboard high definition audio on a pc running running Vista, feeding speakers; and using the analogue output of an Audigy 4 card on a computer running XP, feeding headphones.

Prima facie, a struck triangle is a valid test sound, as the triangle is an instrument of a symphony orchestra. However, could there be something anomalous about the particular triange sample? For example the microphones may have been so close that phase cancellations were occurring. In an auditorium, microphones could be quite some distance away from the percussion section of the orchestra.

I feel like a fish out of water writing on this particular topic. An amateur tredding down a path that others would have investigated years ago! Is there a consensus that a sample rate above 44.1KHz can be beneficial, at least for some musical instruments?

I had always assumed there would be slight differences with a higher sampling rate, but this thread seems to challenge that.

The diffence signal
Another test I did was to subtract version C (original -> 44.1 ->96) from the original 96Kz version (using cooledit). This yielded a difference signal that sounded like a quiet version of the original file. I note that this particular type of test is independent of the precise performance of the sound card used to listen to the difference signal that cooledit computes.

I also subtracted D (original -> 48 -> 96) from the original. The result was not audible at a normal listening gain, despite the fact that when listening to the 96/24 versions separately [version D and the original 96/24 version] I could hear a difference (confirmed with a quick ABX test).

All I can say is that is that's how it reads to me. Perhaps it's just your insults that make your statements seem confused?
Nope, as everyone else knows, I was just wondering if they'd retested those who scored above average.
My, aren't you being especially helpful!

No, I think you're just not reading carefully, or not understanding the concepts involved. I don't see anyone else here claiming to be confused by the two statements.



And you're still wondering, even though, 'as everyone else knows', you were informed days ago that there were no scores 'above average' (at the p<.05 level)?

Btw, your statements seem confused. Are you saying you're NOT suggesting that that a proper re-test of a putative high scorer on a DSD vs Redbook test, would be to see if they could tell 16-bit from 24-bit audio?




You don't seem to have exploited the help you've already been given.

I don't have access to Audition. Is it possible for you to make the resampled versions available for download somewhere. From online tests, it seems as though Audition's resampling is very good. Also, try out the free version of r8brain at 44.1kHz.


I see three possibilities here:

• You can hear frequencies over 20kHz.
• Your soundcard's DAC is doing something different with different sampling rates, possibly performing resampling.
• Your speakers/amp do something strange with high frequencies - like excessive intermodulation distortion.

This makes me think that it's an effect of the resampler, and not your hearing. While it's possible you can hear frequencies above 20kHz, it's very unlikely you can hear above 44kHz. Maybe trying out r8brain is the way to go.

A struck triangle is still a legitimate music sound, wherever it's recorded from.

No consensus I have come across. A lot of people have done tests (like the papers presented earlier in this thread) without statistically significant results.

Thx cabbagerat.

Did some quick tests with r8brain at its maximum quality settings. It performed more accurately than Audition 3. For example, my new file C (96 > 44.1 > 96) could be subtracted from the original struck triangle sample and leave no audible difference signal. [I noted in my post above that Audition 3 produced quite an audible difference signal with its sample rate conversions via 44.1Khz compared with the original sample.]

Am a bit pressed for time so will mention this: I found the r8brain 44.1KHz version did sound slightly duller (with my XP computer, Audigy 4 sound card, and headphones) and this difference was ABXable.

As this could have been due merely to differences in my sound card's filtering on playback, I then opened my new version C file and the original 96/24 sample, in foobar. The converted version still sounded slightly different (as if a tone control had been used to make the converted version slighter less bright). This was ABXable.

I don't have time at the moment to do uploads but may get around to that soon and can then post again.

MMn, that doesn't sound promising.

I don’t know about Audition 3, I still use CoolEdit200. The Help under Convert Sample Type here is quite clear. It says that quality settings of 100 to 400 give the best results. Higher quality settings can cause high frequency ringing because of the steep filters employed. Since this recording has so much energy above 22050Hz, it may be a good candidate for such problems.???

In the old Syntrillium forum, the word from the developer was to use 250 for the quality setting. Calculation times are greater at larger settings but perceived quality will not improve above 250.

What does this mean? If you compared the original with a resampled to 44.1 back to 96, there would have to be a major difference since nothing above 22050Hz could be in the resampled to 96kHz Do you simply mean you could not hear anything from the difference file?

I did not read the help, but simply selected the maximum quality, assuming it would give the best result. There was also an option not to use any filtering at all but that didn't seem a good idea so I left filtering on. In light of this, I guess I'd better read up on what the r8brain help has to say about the quality setting.

Yes, simply that. During playback the volume bars on cooledit showed a burst of signal at the beginning of the difference file, but I could not hear that burst (using the same gain setting as for listening to the unaltered sample). As you suggest, there would have been high frequency content in the 22050 and above range.

I find this technique of listening to a difference file quite useful for pinpointing weaknesses (or anomalies) in digital signal processing. It enabled me to establish that Audition 3 at its maximum quality setting was fractionally altering the overall level of frequencies well down into the human audible range, when perfoming a 96KHz to 44.1KHz and then back to 96KHz conversion.

*************

I suspect that this matter of what filtering to use may ultimately prevent coming to an agreement on the resampling question in relation to 44.1KHz vs 48KHz. Any attempt to present a case that the 44.1KHz version sounds different can be dismissed by reference to filtering effects. The Nyquist limit being relatively close to the upper limit of human hearing, no doubt makes filter design difficult.

Again, perhaps you could post some lossless samples of

1) the original audio

2) the audio after your processing

'
With #1, other people could attempt to replicate what you did (resampling)...with 1 and 2 they could replicate the listening test you did and measure what, if anything, is different.

MLXXX

It'd go a long way if you could please post your ABX results.

Andrew D.

Yes I can and will do that, digital, though at this stage my main interest is receiving suggestions as to what software to use for the sample rate conversions.

I could merely upload the r8brain high quality conversions (which I have informally ABX tested and found to be distinguishable), but if someone can suggest software that will have less effect on the human audible tonality for a 44.1KHz sample rate than r8brain, then I am ... er ... all ears.

One suggestion (AndyH) was a lower quality conversion setting for Cooledit, and I guess I could try that with Audition 3, or with my expired trial version of Cooledit.

I wouldn't trust much signal substraction techniques here, because the filtering may introduce small delays in the processed signals. This delays will leave a small residue when doing the substraction, no matter how good the resampling.

To check if there is such an issue, with Audition/CoolEdit Pro take a spectral look at the difference signal with the FFT view or whatever it was called. If it has significant content (say over -120 dB) only at the ultrasonic part, then both the resampling is ok and hasn't caused any delay either.

AFAIK Audition/CoolEdit Pro resampling is very good, but make sure you have the pre/post filtering option enabled. 400 quality will make a very sharp filter, leading to a possible long ringing at half the sampling frequency (22050 Hz) if the signal has content at this exact frequency. As said, try with 250 to see if this makes a difference.

Inconclusive investigations at a target rate of 44.1Khz:

Quality settings

The wording of the Cooledit 2.00 help on conversion quality includes the following:

Low/High Quality
Use this slider to adjust the quality of the sampling conversion.
Higher values retain more high frequencies while still preventing aliasing of higher frequencies to lower ones, but the conversion process takes longer. A lower quality setting requires less processing time, but results in certain high frequencies being 'rolled off', leading to muffled sounding audio.

I decided to give Cooledit another trial (on an old pc on which it had not previously been trialled). I found that reducing the setting from 999 to 800, 600, 400, 250 and down to 150, merely softened the clarity of the sound more and more, but did not improve it.

The behaviour was similar with r8brain - it sounded best (to my ears) at the highest quality setting. Similarly with Audition 3.0.

Subtracting converted files from each other

Cooledit produced odd results in that it was not internally consistent up to midrange frequencies. A low quality conversion subtracted from a high quality conversion did not merely yield whispers of very high frequencies but a whole swathe of sound well down into audible frequencies.

Both r8brain and Audition 3.0 produced the same level of audio output up to midrange frequencies. Subtractions between r8brain and audition 3 yielded only very high frequency audible sounds, and the sound was faint. I was inclined to reject Cooledit, based on its internal inconsistency.

No proven converter available

My difficulty in proceeding further with this exercise is that the highest quality level conversions of r8brain and Audition 3.0 are yielding slightly different sounds, and they both differ from Cooledit.*

In these circumstances, I cannot draw any definitive conclusion from any positive ABX result when comparing the 96KHz version of the struck triangle to a 44.1KHz conversion using any of these three items of software.

Any difference I heard could be explained away by reference to the filter characteristics used for the conversion.

The only tentative conclusion I can draw is that 44.1KHz may be too low a sampling rate for practical filters. If that is so, then a higher sampling rate may be called for.

I would mention that to my ears there is a greater apparent difference between the original sound sample and any of the conversions (listened to with foobar, an Audigy 4 card, and headphones) compared with differences between the conversions. All of the conversions sound a little duller than the original.

However there could be a large number of reasons why a sound card might perform differently with a 96KHz input than with a 44.1KHz input, including its own filter settings.

As for the differences in sound resulting from use of the three forms of conversion software, I have to assume the filter implementation is different in each case, and this is giving a slightly different colour to the processed sound.

______________________

* ABX report for two of the converters, r8brain 1.9 and Cooledit 2.0, operating at their highest quality settings and converting a 96Khz/24 bit file to 44.1Khz/24 bits:-

foo_abx 1.3.1 report
foobar2000 v0.9.5.1
2008/05/06 04:02:33

File A: C:\Documents and Settings\All Users\Documents\triangle-2_2496_r8brain-conversionTo44-1--HighestQuality.wav
File B: C:\Documents and Settings\All Users\Documents\triangle-2_2496--Cooledit--ConversionTo44-1--quality999.wav

04:02:33 : Test started.
04:02:54 : 01/01 50.0%
04:03:48 : 02/02 25.0%
04:04:53 : 03/03 12.5%
04:05:28 : 04/04 6.3%
04:06:41 : 05/05 3.1%
04:06:43 : Test finished.

----------
Total: 5/5 (3.1%)

And now a similar ABX report comparing the conversions of Cooledit 2.0 and Audition 3.0 with each other:-
(Note: these two sounded quite similar and were not easy to ABX!)

foo_abx 1.3.1 report
foobar2000 v0.9.5.1
2008/05/06 19:48:04

File A: \\action\shareddocs\triangle-2_2496--Audition3convertingTo44-1KHz-quality999.wav
File B: \\action\shareddocs\triangle-2_2496--Cooledit--ConversionTo44-1--quality999.wav

19:48:04 : Test started.
19:48:55 : 01/01 50.0%
19:49:53 : 02/02 25.0%
19:51:59 : 03/03 12.5%
19:52:39 : 04/04 6.3%
19:53:00 : 05/05 3.1%
19:53:23 : Test finished.

----------
Total: 5/5 (3.1%)

LINKS TO THE THREE CONVERSIONS:
Audition 3.0 version: http://www.hydrogenaudio.org/forums/index....ost&id=4441
Cooledit pro 2.0 version: http://www.hydrogenaudio.org/forums/index....ost&id=4442
R8brain 1.9 version: http://www.hydrogenaudio.org/forums/index....ost&id=4444


LINK TO THE ORIGINAL SOUND SAMPLE:
The original 96KHz/24-bit sample of a triangle being struck can be located on the excellent PCABX test page: http://64.41.69.21/technical/sample_rates/index.htm
The relevant sample is the one marked "Triangle Reference Presented At 24/96".

Sure, go ahead, think whatever you like. I see that you sure like telling us what that is! As for anyone else, why would they want to get involved in your argument? Personally, I simply don't understand what you hope to gain by insulting people. However, if you feel it helps your cause, please, don't stop simply on my account.

As I said, I "was" wondering. Perhaps you should try reading my post more carefully?

There you go with the insults, again. Good luck with that!

Why don't you quote me?
I'm doing just fine, thanks. By the way, I feel I must compliment you on how well you've chosen your nickname. I found it amazingly appropriate.

Thanks!
Hopefully, I'm not just beating on a dead horse here, however, I do have several more questions, please? First off, could conducting audio tests at higher than normal listening levels, reveal subtle differences being missed by current ABX testing? Secondly, doesn't the one 20/20 score in the Detmold study, merit further investigation into exceptional cases of hearing ability? Third, is there any possibility that the test equipment was simply unable to reproduce the difference?

Furthermore, I feel I must apologize in advance if these questions seem too repetitive or rudimentary for some people here.

If there is one person way outside the range encompassing everyone else, even if that one person’s score is completely valid, we have to ask if the fact has any relevance. Suppose one person in a million can really detect a difference, but the other 999999 can not? If you happen to be interested in the abnormal, then you may want to located these (relatively) few individuals so you can subject them to laboratory degradation, but if you are interested in just about any other aspect of audio, you probably could not care less about them.; they just are not relevant.

There is a possibly important aspect of the test equipment in such comparisons. Many, possibly most, soundcards have somewhat different performances at different sampling rates. If no one detects any difference in the audio, the soundcard differences probably don’t matter, but if there are positive scores, we have too many variables to eadily determine why. at the very least we need to repeat the tests with different, high quality, DACs.

With higher sound pressure levels, more intrinsic audio differences will be audible. This isn’t specific to different sampling rates, it is a normal part of every day sound. Suppose something is audible (only) at very high levels. The basic question of paragraph 1 applies. Do we care? Why do we care? Will the fact ever be relevant at any time other than during such a test?

The equipment again comes into consideration. It is possible to build enormously powerful amplifiers, but the transducers for converting that electrical power into sound are another matter. Speakers without a lot of distortion get to be very expensive. I think it is probably not possible to brush aside the strong possibility that because of complex interactions in the transducers, higher frequency distortions might have effects on audible frequencies that would not occur otherwise.

A word of caution here: the test is being passed with an on-board sound card (and a Creative sound card) and unknown other equipment.

If you let me pick the sample and sound card, I too will pass the test, and I can't hear a thing above 17kHz!


It would be interesting to know what is reaching MLXXX's ears. He implies that A, B, C and D are so bad there's no need to ABX, and then ABXs "E" with ease.

Whereas I am not aware of a single published test where someone has ABXed a 24kHz low pass filter, unless something in the signal chain has been effectively broken.

Either we are conversing with the bionic man, or his audio equipment is faulty.

How to tell? It would be a start to try recording the analogue output of the soundcard using a decent sound card - and see what differences are present between the 48kHz, 96kHz, and 192kHz versions.

Cheers,
David.

David, I can understand your concerns. However I have now edited my post above (#72) to include links to the three files.

Anyone can download these and do their own ABX tests. Note that as the files are all at the same sample rate, 44.1KHz, whatever sound card (or AVR etc) is used to compare the three files, will invoke the same filter for each.

Regarding my own hearing of high frequencies, it is not exceptional, one reason being that I am middle-aged.

___________________________________

Edit: I have now tested my hearing using headphones connected to the output of an Audigy 4 hub (an external sound module manufactured by Creative that interfaces with a PC). In the ABX test below, the purpose of the 40KHz sample is to serve as a reference. I found that at a normal gain setting for listening to music, I could perceive the test frequency up to about 19KHz fairly easily, albeit that that high frequency was very faint.

At 20KHz there was no perceivable tone (just some low level sound card noise). When I tried 19.5KHz, I could just hear the tone again.

I recorded the test tones with Audition 3.0 at a 96KHz sample rate, and at a level of -20dB. Here is the ABX result:

foo_abx 1.3.1 report
foobar2000 v0.9.5.1
2008/05/07 00:03:29

File A: C:\Documents and Settings\All Users\Documents\sine19500@96KHz.wav
File B: C:\Documents and Settings\All Users\Documents\sine40000@96KHz.wav

00:03:29 : Test started.
00:03:47 : 01/01 50.0%
00:04:05 : 02/02 25.0%
00:04:14 : 03/03 12.5%
00:04:32 : 04/04 6.3%
00:04:48 : 05/05 3.1%
00:04:51 : Test finished.

----------
Total: 5/5 (3.1%)

MLXXX,

Thanks, but that's not quite what I meant. I know pretty much what Cool Edit / Audition will do to the file.

However, I don't know what your OS, drivers, sound card, amplifier, and speakers or headphones do to the signal.


If this was a real experiment, I would put a high quality probe microphone in your ear, and compare what it picked up from the 96kHz version vs the other two. I would also re-digitise the output from the amplifier, and the output from the sound card. I would analyse all three sets of recordings to see if/where distortion crept in.

From your description, it seems very likely that the audible difference is well within the audible(!) range when it reaches your ears, and as such is being generated within your equipment.


Since you have two sound cards, would you be willing to try a further experiment?

Get one to record (maybe the creative one - will it record at 96kHz without conversion anywhere?) and the other to playback. If possible, play back the 96k, 192k, 48k, and 44.1 on the other PC, recording them all at 96k. Then post the result. Also, listen to the result and report what difference, if any, you hear between the various re-recordings.

Cheers,
David.

That strikes me as quite a project to undertake, with many variables.

At this point in time I would be pleased to hear reports from anyone inclined to ABX compare the three examples of conversion to 44.1KHz I have uploaded. If others can hear differences too, then we appear to be on a sticky wicket as far as undertaking more advanced tests is concerned, such as comparing a 48KHz file with a 44.1KHz file.

Ok, I had a look and a listen at the files you presented. First things first, I failed to ABX any of the samples - but that doesn't prove anything - the speakers I used don't work too well above 17kHz anyways*. Second, your resampled versions are 32bit float, and the original is 24bit - are you sure your ABX software or soundcard are not treating these two depths differently? The third point is that there is obvious ringing present on the left channels Audition and Cooledit versions (in a plot of the data) - mostly just before the sound starts - this is not present on the R8brain or original samples. Level matching doesn't seem to be an issue here - the Audition, R8brain and original versions all match very closely on A weighted level. Anybody with ABX success?

* ABX Setup: AV-710 rear DAC into a LM3886 based chipamp driving full range speakers with Fostex FE167E drivers.

Thanks for your observations, cabbagerat.


I could not find an option in Cooledit or Audition to create a 24-bit result when converting the sample rate. I therefore selected 32-bit float. At the last moment when uploading the files, I realised someone might pick me up on the fact that the r8brain conversion was at 24-bit and not at 32-bit float. So I quickly redid the r8brain conversion so it would match the format of the other two conversions.

I cannot hear any difference between the r8brain 24-bit version and the r8brain 32-bit float version. However, as you say, for some reason during playback my setup might have reacted differently to a 32-bit float version, compared with a 24-bit integer version, so it was as well to check this out.

As matters stand, all conversions are in the same bit format (32-bit float) and at the same bit rate (44.1KHz).

I did find the Audition and Cooledit versions extremely difficult to tell apart with headphones (and the Audigy 4 hub). I found them slightly easier to tell apart on my main hi-fi setup and listening to my AVR doing the digital to analogue conversion, and sending the audio to speakers.

As the files contain so much high frequency energy they are quite fatiguing on the ears, I find. It is partly for this reason I stopped at 5 tests in the ABX testing, as the doing of the testing was affecting my ears. Even some hours after the ABXing last night, my ears were still ringing slightly and feeling a little uncomfortable. I used normal listening gain, but the samples do contain an extraordinary amount of high frequency energy.


The ringing in the left channel might explain why the stereo image appeared to me to shift to the left in the Audition and Cooledit conversions, compared with the original 96KHz file.

______

1. Anyone else with comments on the three conversions?

2. Is there some other downsampling method that might leave more of the original sound intact?

3. If the need arises to upload a file illustrating another conversion, there is now an upload thread titled Resampling down to 44.1KHz, Is there a method that will not colour the sound? here.

EDIT:
On further reflection I wonder whether I have reached the point of hijacking this thread towards an overspecialised technical discussion, better suited to a standalone topic. I'd be happy if people interested in discussing the three files further do so in the related upload thread, namely: Resampling down to 44.1KHz, Is there a method that will not colour the sound?

MLXXX,

I'll say it one last time. It is almost certain that your playback is colouring the sound, while the resampling is fine.

Which is why, unless you can try the experiment of re-recording the output of one sound card with another, this is a waste of time.

Or, to put it more simply, there's no point carrying out an experiment without first checking that the equipment is as described, and working correctly.

Cheers,
David.

Personally. I'm not interested in either the abnormal nor the extremes, I'm simply interested in what I may be able to appreciate. Perhaps others cannot, but I'm not interested in a numbers game, nor in statistics per se. What interests me is in seeing how far I can go in improving the quality and usability of my music collection.

Please, don't get me wrong, scepticism is a good thing, however, clearly closed-mindedness is not. I, myself, am simply trying to be open-minded, honest and inquisitive, nothing more. I'm not sure but it seems to me as if some people seem to think that if it isn't audible to a certain percentage of ABX testers, it doesn't exist. In truth, ABX testing is wholly acceptable as a method of scientific enquiry, however, how the results are being interpreted is still somewhat subjective, in my humble opinion. Now, I believe this is necessary since each subject must define what is a personally useful level of audio high fidelity. I'm simply seeking what works for me. If you have found what works for you, great! Perhaps through sharing and discussion, it will be possible for people in general to find greater joy in their own personal music experience, no?

Normally, if three files of identical bit depth and sample rate are played back using the ABX feature in foobar but they sound different, and the ABX statistics confirm the probability of the result, then we conclude they are in fact different, by the standards of this Forum.

You have asked me to perform and upload recordings. That seems to me to be an unusual approach which could raise as many questions as it might answer. I would prefer it if you yourself or perhaps some very young members or others with very good high frequency hearing were to ABX the three files I have provided. This would be simple and direct.

It may be worth stressing that at this point I am not comparing the 96/24 clip with the conversions. I am merely comparing the three conversions with each other. Those three conversions are all at 44.1KHz 32-bit floating.

Edit: I might add that there were three different DAC devices available to me and all three gave perceptible differences as between the three files, namely: the DAC in my Audio Video Receiver, the high-definition DAC in the motherboard of my Home Theatre PC, and the DAC in the Audigy 4 external sound module.

Accepting the results of group ABX testing makes perfect sense to me if the individual is prepared to accept that they are highly likely to be "normal". Performing individual ABX testing allows you to draw your own conclusions regarding your own individual hearing abilities. So what's the problem?

Cheers, Slipstreem.

2tec, my understanding is that a good standard of equipment was used, with a large sample of people listening at normal listening levels, to a variety of music. If there had been a clear difference for even a small percentage of participants, that would have emerged from the testing.

What I do not know is what parts of the music were put through a 44.1Khz sampling rate bottleneck. Unless the change occured right in the middle of particularly sparkling passages full of high frequency energy, it does not surprise me that the temporary bottleneck went unnoticed.

From my own investigations and reading, I believe that a reduction in sample depth to 16 bits is only identifiable if dithering is poor or the listening level is unrealistically high. That to my mind leaves sample rate as the significant factor in a 'bottleneck'.

Perhaps people more familiar with the report can comment on what passages in the music got the 'special treatment'. Was it by any chance a series of clashes of cymbals with a quick change to 44.1Khz in the middle of that series? Unless such 'killer' episodes were included, identification of the bottlenecks could be expected to have been quite difficult, all other things being equal.

If I really had written or even implied something as foolish as 'nobody can hear better than anyone else', as you claimed I did, it's a fair bet that someone here besides you would have taken me to task.

As for your pose as the blameless victim of mean old krabapple, it might play better if you revised all your posts on this thread. Doubt it, though. Btw, that part where you thank Pio for reminding you of what *I* wrote: priceless.




To nail this part down, perhaps someone can supply a couple of files of demonstrably first-rate resampling (every step approved by HA noggins), and MLXX can see if he can ABX them? Then at least we can narrow it down to either, his extraordinary hearing, or his sub-extraordinary playback chain.

In the meantime, maybe I can get Arny Kruger to take a look at the thread, and offer some opinions.

Sounds good.

Sorry MLXXX, I wasn't referring to the ABX test itself - your other thread is subtitled "Is there a method that will not colour the sound?" - that implies a comparison with the original, which to you yields a clearly audible difference. It's that comparison that I was questioning, and that comparison that I suspect your playback is colouring.

You should do a search for the discussion of "udial"...
http://www.hydrogenaudio.org/forums/index.php?showtopic=9772
... to see why this is can be anything but simple for samples that contain lots of HF/ultrasonic energy.

It's still an interesting test, but you have to be in possession of all the facts, and wary of the potential pitfalls.

Cheers,
David.

Yes there are pitfalls everywhere with 44.1Khz. These include:

(i) the digital filtering for the analogue to digital process required to capture an analogue source at a 44.1Khz sample rate
(ii) the digital filtering in the digital to analogue conversion required to play a recording made with a sample rate of 44.1KHz
(iii) [allied to (ii)], the fact that there may be an intermediate resampling to a card's 'native sample rate'.

Digital filtering must find a compromise solution that:
* provides adequate protection against aliases
* maintains the frequency response up to a zone not far below the Nyquist frequency
* avoids excessive phase changes, or ringing, or other distortion.

Why would (i) or (ii) be any different for 44.1kHz than 48kHz or any other sampling frequency for that matter?

Because there is very little headroom. The 44.1KHz Nyquist limit, 22050Hz, is awkwardly close to the upper limit of human hearing, of around 20000Hz.

I forgot about udial -- couldn't MLXX use that to see if his setup is resampling during playback?

Here's udial.flac

Yes, and the ATH curve drops off severely as it approaches 20 kHz. If only our ears respond like studio monitors...

It's remarkable that most recording- and mastering engineers who use (and claim audible advantages of) 96 kHz and higher sample rates, are still using rather standard bandwidth microphones and monitors (upper limit slightly over 20 kHz). Audible benefits of hi-res audio (QED) should therefore probably be searched for in the audible band (<22kHz).
Please note: 96 and 44.1 kHz versions can sound different, but this doesn't necessarily mean that the 96 version sounds better. It could be worse ! David Griesinger pointed out in this paper that high-frequency content can cause InterModulation distortion in the playback chain. It is possible that a LowPassed version reduces or eliminates this effect and therefore sounds better on some playback systems.

I have created an account only to post this. While a lot of helpful information is available on the forums, access to it didn't need my input. However, some people here tend to be strongly biased.

I have pretty good hearing, probably also due to my age (17). I can hear sounds as high as 23 kHz and even 24 kHz, however extremely faint. I am certain of this values, they are quoted from a medical examination and not some cheap speakers. In addition, I have Asperger's, and one direct effect is the ability to abstractize and categorize sensory input, including sound. I can clearly distinguish every instrument type in a symphony, for example.

I also do not care about the rest when it comes to sound quality. Whether they distinguish a 128 kbit/s MP3 from a SACD is not important to me, what matters is that I do. I had the occasion to compare classical music in ABX between a SACD and a version resampled to CD-A quality and I could with fairly high precision identify them. Another facet of Asperger's is that I have synestezia and I perceive some sound combinations as emotions and tastes rather than abstract vibrations. With a high degree of subjectiveness, I have found the SACD to convey a sense of serenity and trippy tranquil that the CD-A did not, to the same degree. I assume (but am not sure this is correct) that the higher harmonics were the cause for this.

I own a fairly cheap stereo at home and while it does sound clear, it peaks at about 17 kHz. When input is 20 kHz, output sounds more or less like 5-10 kHz for example. So I cannot enjoy music fully and do not afford anything more than, say, a thousand euro, which I assume is well under the price tags for pro audio.

Well so, this was my opinion on the subject. I only wished to change some biased opinions in that there are some people who can tell the difference and who most likely are among those who say CD-A is not enough. I also need to improve my non-native English, as some phrases do sound awkward...

Anyway, thank you for reading!
gantrithor

PS: By the way, I happen to have received a present consisting in a "Millenium Masterpieces" collection a few years ago. The box has the inscription "20 bit recording (DDD)" on it multiple times. I knew CD's are usually 16bit/44.1kHz/Stereo but thought 20bit is also possible (perhaps more throughput, less duration). I suspect the recording was made at 20bit, then downsampled to 16bit on CD mastering. But why do such a thing? Any ideas? Thank you again!

20 bits is useful for noise shaping. You can get >96db of dynamic range at frequencies the ear is most sensitive to (midrange/treble), in exchange for <96db at less sensitive frequencies (very high treble).

Have you considered scoring some headphones? You can certainly get good ultrasonic response if you know where to look. For 1000 euros you can get pretty much top-shelf headphones with change to spare for a good amp. The Sennheiser HD650s are "only" 450 euros, and they arguably have a good response out to 30khz. Hell, you could buy a used Stax electrostatic rig at that price.

It's not uncommon for teenagers to be able to hear up to 24kHz at high levels. Some individuals have surprisingly low thresholds. I've attached a graph of some averaged results.

See this paper for the actual results:

Henry, K. R.; and Fast, G. A. (1984).
Ultrahigh-Frequency Auditory Thresholds in Young Adults: Reliable Responses up to 24 kHz with a Quasi-Free-Field Technique.
Audiology, vol. 23, pp. 477-489.

The response at those high frequencies drops off due to noise exposure before the normal audiometry range (typically only measured up to 8kHz) shows any change.

There's some fascinating research in this area. However, few (if any) people believe that reports of audible differences between CD and other formats have anything to do with high frequency hearing.

Cheers,
David.

P.S. There are responses at 40kHz-50kHz via bone conduction. That's a whole separate topic!

Improperly designed filters may interfere with signals far away from their roll-off frequency. The simpler (~cheaper) the filter the more likely it is to affect what it should not. The most problematic is the use of closed-loop (IIR) filters which tend to have nonlinear phase characteristics but provide steepest response and shortest start-up for the chosen order (~price).
You may not even have an idea about how many such filters has the sound actually passed between the microphone at studio and the loudspeaker/headphone at your home.
And I did not even mention the frequency/phase characteristics (~deformations) of microphones/amplifiers/cables/loudspeakers and whatnot which are unintended filters as well.

96+ kHz rate helps the software/hardware design (~reduces cost to achieve comparable results) as the "signals far away from their roll-off frequency" are much much farther away than with 44 kHz.

I think that 44 kHz is perfectly able to capture human-audible content. However, the practical results are plagued by the mentioned design/cost limits.

We keep coming back to the conclusion that 44.1/16 is perfectly adequate for distribution of the final product due to limitations in storage/bandwidth, but for any other use there are practical advantages to higher sampling rate and bit depth.