When is hi-rez overkill?

[Almen]

I could say that I feel sorry for you. A higher quality format that you are not able to appreciate, but that would be condescending, no?

Well, luckily I appreciate all my HDCDs, HDADs, SACDs and DVD-As very much, so it's not an issue.

 

[G E]

"When is hi-rez overkill? "

A friend of mine mentioned this thread a few days ago and I decided to look through it. Luckily it is only 4 pages give or take a thread so i did not waste a lot of time.

I don't believe there is to be an amicable resolution of the differences of opinions here. I don't think it matters either.

Like many people here I want the best possible sound reproduction I can afford and the best produced software of music I cherish. I don't have equipment to measure distortion or frequency response of my speakers in my dedicated listening room but if I did and had to choose between spending time measuring stuff or listening to music, I choose the latter. But I understand the interest in measurements and subjective tests.

I have performed amplifier shoot outs in my home when helping my wife choose her system. Three amplification systems by three very well regarded manufacturers with three completely different sonic signatures: Unscientifically we noted "air", "voicing" and "imaging holography" among other attributes. The differences between amps were not small and since we had no equipment to measure things we relied on our hearing to tell us what we liked best. This was not a blind test, I had a favored brand, and no it did not win the shootout.

Hi resolution will not solve bad mastering- Exhibit A: Plant/Krauss' "Raising Sand", awful sound in all of its formats CD, hi-res and LP. Too bad, it's a great performance of some very good songs. Maybe the MP3 sounds better because they throw away so much of the music. Haven't tried that.... I'll let someone else go first.

My favorite recordings in the digital domain tend to be 96/24, they posess a creamy silkiness (or is it silky creaminess?) that just isn't there on 48/24. And, in addition, I find SACD to be more satisfying than 96/24 dvd-a. Well done CDs can be satisfying too but I always play them *before* bringing out the hi-res stuff. Otherwise it's like listening to music with a head cold. After all, CD's sampling rate is much cruder than the higher resolution formats and it has a brickwall filter in the low 20 kHz's. Of course, all my observations are in an uncontrolled environment, with different recordings enjoyed with different brews. But the experiment has been repeated *many* times.

Last week I had the ultimate high resolution surround sound experience: I attended a live performance of Brahm's German Requiem. 10th row, slightly left of center. I didn't play my record player when I got home.

 

[Almen]

My favorite recordings in the digital domain tend to be 96/24, they posess a creamy silkiness (or is it silky creaminess?) that just isn't there on 48/24. And, in addition, I find SACD to be more satisfying than 96/24 dvd-a. Well done CDs can be satisfying too but I always play them *before* bringing out the hi-res stuff. Otherwise it's like listening to music with a head cold.

You are also mistaking the format for the mastered product. Unless you are comparing two identical sources (as described in the aforementioned report) this makes no sense.

SACD has less resolution than CD above 7 kHz! So it is not hi-rez you are preferring when favouring SACD/DSD before PCM.

Last week I had the ultimate high resolution surround sound experience: I attended a live performance of Brahm's German Requiem. 10th row, slightly left of center. I didn't play my record player when I got home.

Well, that's the highest resolution you can get. I too have a ticket to "Ein Deutsches Requiem" later this fall.

 

[G E]

Well, that's the highest resolution you can get. I too have a ticket to "Ein Deutsches Requiem" later this fall.

I see your location is Sweden. If your ticket is for the Atlanta Symphony Orchestra Chorus with the Berlin Philharmonic you are in for a treat! Concert dates are December 17-19.

 

[G E]

You are also mistaking the format for the mastered product. Unless you are comparing two identical sources (as described in the aforementioned report) this makes no sense.

That is correct. I am not comparing two identical sources. I am not measuring. There are no double blind sampling sessions. I am not running a scientific experiment. I don't want to when I listen to music.

But I do know what music makes me happy or takes me on an emotional journey and that is what is important to me.

 

[ssully]

That is correct. I am not comparing two identical sources. I am not measuring. There are no double blind sampling sessions. I am not running a scientific experiment. I don't want to when I listen to music.

But I do know what music makes me happy or takes me on an emotional journey and that is what is important to me.

Yet you could not resist adopting scientific terms to imply that what seemed different to you, really was different: "But the experiment has been repeated *many* times."

A badly controlled experiment doesn't get better the more times you repeat it.

 

[Almen]

I see your location is Sweden. If your ticket is for the Atlanta Symphony Orchestra Chorus with the Berlin Philharmonic you are in for a treat! Concert dates are December 17-19.

Actually no, I live in Gothenburg and it is the Gothenburg Symphonic Orchestra and the Gothenburg Symphonic Choir conducted by Christian Zacharias.

I guess it'll be OK anyway.

 

[simonwagstaff]

Yet you could not resist adopting scientific terms to imply that what seemed different to you, really was different: "But the experiment has been repeated *many* times."

A badly controlled experiment doesn't get better the more times you repeat it.

I think he is using his "goosebumpometer", a very sensitive and accurate instrument when used properly, but has low precision. Measures all aspects of audio reproduction as it relates to the listener but with too many variables to ensure accuracy in repetition.

I think you should pick one up anyway!

 

[G E]

I think he is using his "goosebumpometer", a very sensitive and accurate instrument when used properly, but has low precision. Measures all aspects of audio reproduction as it relates to the listener but with too many variables to ensure accuracy in repetition.

I think you should pick one up anyway!

Wagstaff has grasped the essence of my brief essay: That which creates goose bumps is good, desirable and should be repeated.

But it isn't always replicated. Some nights the music flows and it is a thrilling experience. The next weekend it does not reach the same heights even playing the same recordings. The system has not changed. Something has changed in the way *I* am processing the events.

Bad science? Absolutely!

Or maybe the "goosbumpometer* needs recalibration

 

[Key]

Hmm this is something I have been experimenting and thinking about lately - if you really need to go above 44.1kHz 16-bit for the delivery format. Honestly I don't know where I stand on the subject. But I tend to think there are a lot of variables in this problem that can effect the outcome of any comparison. What Dither is being used? What SRC is being used? Was the dither done prior to the SRC or the SRC prior to the dither (this can have a big effect ime and I am now thinking the normal order that pros preach as being correct is actually the wrong way to go depending on what dither is used). I think the characteristics of the output DAC and if you upconvert before the output will have a larger effect because of the DAC output filter and how it will effect frequency response and phase response.

 

[ArnoldLayne]

I take the liberty to revive this discussion with a link that was posted elsewhere in the forum:

http://www.avguide.com/forums/blind-listening-tests-are-flawed-editorial

In short, a blind test concluded that no quality differences were detected between uncompressed audio and a certain lossy codec. Afterwards, an expert easily identified an audible artifact for the very same codec. I haven't been able to check article sources but let's just assume that these things can happen. Which illustrates a fundamental limitation of ABX testing: failing to prove that A is different from B is not necessarily a proof that A and B are equivalent. But don't shoot me just yet, I admit that a difference between A and B identified via long-term home listening is not necessarily conclusive either. In both cases, a fundamental mistake seems to often (if not always) be made: the "experiment" is designed in advance and it's aims are defined ad hoc afterwards. E.g. we set up the gear, do the listening, succesively build notions about differences and finally draw some statistical conclusions. But if we don't know exactly what we are investigating, we cannot avoid bias. Whatever the result, doubts can typically be put forward about the influence of everything from DAC design to hearing fatigue. So I think it would be better to return to a more orthodox method and state the hypothesis before testing it. As starting point for a possible theoretical framework I would like to quote a piece from the discussion of the mentioned article:

Third, there's been a suggestion that although we can't hear sinewaves above 20kHz, we can detect the steepneess of transient signals that implies a bandwidth greater than 20kHz. We use these steep transient in localizing sounds. Part of the HDCD process (the patent application makes for fascinating reading) encodes in the hidden information channel indicators that a signal's transient leading edge is less steep on the 44.1kHz/16-bit signal compared with the high-res original from which the HDCD-encoded compatible signal is derived. A conjugate process in the decoder restores the transient's orignal rise time.

I have understood that "transients" are waveforms, typically at the beginning of a tone or word, without any particular shape in terms of frequency. More generally, we can think of many situations in music when the sound waves' amplitudes and phase are not constant between two sample points. Intuitively hi-res would increase the accuracy since the samples cover shorter time intervals. Does anybody have more information on this, has the issue been covered somewhere? I have searched for it in many places, but cannot find any good articles about it.

Thx in advance
AL

 

[ssully]

Experiments need to be designed in advance; this is not a 'mistake'. It's good method. As is DBT, which not 'unorthodox' in the least, in sensory testing.

That codec test Harley recounts was done in *1991*. The virtual dawn of lossy codecs. Would they be applicable to , say, the widely-heard LAME mp3 or AAC codecs of 2009? Not to mention that I wouldn't trust the highly anti-DBT Harley to get the details right. Locanthi died in 1994, Harley is recalling, from memory in 2004, a taped talk by Locanthi from a 1997 AES workshop.

Bart Locanthi was a strident anti-lossy crusader from the start. He was also AES President for a year or two in the late 1980s. To twist his stance against lossy as being somehow a stance against DBT generally, is not supported by Harley's anecdote.

I'm downloading a couple of the Grewin papers from the AES site about the Swedish lossy codec trials, to see if more accurate light can be shed on them.

As for the 'transient' stuff, it sounds like the old canard about Redbook being unable to resolve timing differences that are 'faster' than 1/20,000 of a second. In fact the time resolution of Redbook is in the picosecond range -- that's trillionths -- as e.g., Bob Katz demonstrates in his book 'Mastering Audio'.

 

[Almen]

I have understood that "transients" are waveforms, typically at the beginning of a tone or word, without any particular shape in terms of frequency.

Well... yes and no. Any signal can be analyzed as a sum of frequency components, even transients. But when looked at in the time plane, a transient is as you say, the onset of a tone, or a very short sound. Typically in music, a snare shot.

More generally, we can think of many situations in music when the sound waves' amplitudes and phase are not constant between two sample points. Intuitively hi-res would increase the accuracy since the samples cover shorter time intervals.

Intuituion is not always correct when it comes to complicated matters. As you say, the first question asked is "What about the things that happen between the samples?" And luckily, it turns out that both theoretically (Shannon/Nyquist) and practically this is not a problem since a filter is used that implements the sinc function (most often used) to the original signal. This means that every little nook and cranny on the original (bandwidth limited) signal ends up on the CD, and is replayed by the D/A converter.

Moreover, the applied dither makes it possible to increase the resolution in the amplitude so that it is better than the theoretical "bit depth" resolution.

I found some information about this in the AES paper (although the author makes some unsbstantiated claims regarding hi-res that is refuted in the paper I referred to previously) "Coding for High-Resolution Audio Systems", J. Audio Eng. Soc., vol. 52, No. 3, 2004 March, by J. Robert Stuart:

Provided that both the correct level of TPDF dither is used in the quantizer, and the signal has no content above the Nyquist frequency (half the sampling rate), then the system has infinite resolution of both time and amplitude (see the worked examples in [15]).

[15] S. P. Lipshitz and J. Vanderkooy, “Pulse-Code Modulation—An Overview,” J. Audio Eng. Soc., vol. 52, No. 3, 2004 March, pp. 200–215.

I repeat: the system has infinite resolution of both time and amplitude. Couldn't get much better, eh?

 

[Almen]

...a fundamental limitation of ABX testing: failing to prove that A is different from B is not necessarily a proof that A and B are equivalent.

That is a fundamental in science as a whole - you can never prove a negative, or an absence of something (the null hypothesis). Nothing strange with that, and nothing specific for ABX testing (not sure if you really mean DBT, here).

The fundamental limitation of a test involving humans (listening, wine tasting, medical tests, or the like) that is not performed blind is that you cannot prove anything.

What you can do with double blind listening tests is give enough evidence for making a null hypothesis highly likely.

 

[Key]

Also it seems as if you think ABX testing and long term home listening are mutually exclusive ArnoldLayne. This isn't the case if you own a PC. Foobar2000 has an ABX feature built into the media player that you can use at your own leisure. You can ABX in your own listening environment and take as long as you like. These tests do not have to be totally blind either - you can know what you should be hearing or looking for in a difference. I do understand what you are saying though - a person who doesn't have a clue as to what they should be listening for to find the difference isn't likely to find it imo. Especially with really subtle differences that can be measured but not easily heard.

 

[Almen]

Key: You're absolutely right. A common misunderstanding regarding blind tests is that it is "a stressful, unnatural situation where the purpose is to trick the participants".

The blind testing itself should be preceded by thorough open listenings where the listeners together discuss differences and artefacts that might be heard.

The fact that some people find it stressful not to hear the things they thought they heard so clearly in open listening is another matter. I find it interesting.

 

[ArnoldLayne]

An ABX with Foobar on my PC would actually reveal significant differences between CD and Hi-res at a very high level of confidence This is because depending on the sample rate, in my setup there are different and very noticable colorings of the sound. It's a basic system: Realtek AL888 onboard sound card and cheap Logitech 2.1 speakers. Despite turning off all DSP effects and such, there's still some manipulation that makes the test useless in this particular case, on my gear that is. Which illustrates what I'm trying to explain about both (ABX/DBT and "home listening") methods: 1) There are hidden factors that can affect any listening test. 2) Lack of predefined hypothesis makes it more difficult to identify and eliminate the relevant factors. This is not only common sense, it is also common scientific methodology sense. But please bear in mind that it's not an argument against ABX tests itself but rather an observation on how they should be conducted.

Regarding the theoretical framework, I truly appreciate your efforts to explain the different aspects of applied signal theory. But I'm afraid I still have this hunch that you may be missing a couple of points. Firstly, the attempts to increase resolution and timing with for example dither seems to me very much like substituting the unpredictable and not possible to sample with something predictable and possible to handle within the model. Secondly, I suspect that the Nyquist theorem holds if and only if the time function is invariable (same frequency content and amplitudes) during some time interval (I've even seen some mentionings about this in Wikipedia for the underlying maths, although I cannot remember where). Putting these two things together, I end up with the feeling that in order to make digital sampling work we filter away the finest and unforeseeable details in the sound. In other words, we assume that music only contains what can be handled with our samples and then filter away whatever content that could eventually be beyond it's scope. Whether or not this hypothetic content is relevant or not for the listener is naturally a different story. As I said from the beginning, this part is about the theoretical framework.

I'm of course not expecting to convince anyone about anything with this layman's talk of mine. As a matter of fact, I would be happy enough if I have made you understand my point of view on the Shannon-Nyqvist based signal theory. As with the ABX method above, I'm not at all questioning the theorem itself. It's merely about some of the conclusions we draw from it.

 

[Key]

Well you should consider the entire chain and measure it thoroughly for this type of test. The closest I have come to a simple conclusion to this type of problem is to basically measure the DAC/ADC loopback on the soundcard I am using. Measure it in every possible setting you can - most likely you will want to stick with ASIO and avoid direct sound. Find which setting measures the best for your soundcard and just resample at 24-bit to that. I am pretty sure the weak link in any test like this will be the analog stage and the output filter for the DAC though and not really the format.

 

[Almen]

1) There are hidden factors that can affect any listening test. 2) Lack of predefined hypothesis makes it more difficult to identify and eliminate the relevant factors. This is not only common sense, it is also common scientific methodology sense. But please bear in mind that it's not an argument against ABX tests itself but rather an observation on how they should be conducted.

I agree with you, here. It is very important to beforehand define "What are we testing?", as well as eliminating any factors that could affect the value of the result.

...I end up with the feeling that in order to make digital sampling work we filter away the finest and unforeseeable details in the sound.

Hm... I don't know what to say, really. For me, it was a great revelation, nay, epiphany, when I read about the sinc function, and how it is used to filter the analog signal that shall be sampled. In this way I understood how, in practice, Shannon/Nyquist is implemented.

Understanding Shannon/Nyquist was not enough for me, so I recommend literatue about A/D and D/A conversion. For instance, JAES has a lot of articles about this.

The second thing to stress is, as you say, how dither is used to increase the resolution. Also, I must recommend the available literature, since forums on the net is maybe not the best tool for complex technical and scientifical problems.

Do you think that any of your questions could be satisfactory answered here?

 

[Key]

Subjective listening aside I have found that if you use a dither like Izotopes which uses ultrasonic filtering/noise shaping *before* you downsample that you will incur much less noise when going down to 44.1 16-bit from hi res. Basically the dither noise is hidden above the nyquist frequency of 44.1 and then filtered out when you downsample. If you downsample before you dither and convert the bitdepth then the dither has no place to go which isn't audible and is forced into the 10-20kHz range.