Home |
Search |
Today's Posts |
#1
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
First of all, let me say that I'm not so concerned any more *why* I
like tube amps and analog, and it certainly could be the distortion. Recently I've been enjoying my iPod played into in-the-ear monitors. I've got a modest IEM, the Shure E2C, but the sound is still musical and very detailed, effortless detail. I've been enjoying it---and the d/a and headphone amp in the iPod would *not* be the sort of thing I think of as high-end, although they may very well have excellent specs. Anyway, on the topic of blind tests, I have a concern about them, but I will state the concern and give y'all a chance to address it. Let me say first that I'm NOT using this as a justification for super-high-end amps and cables and cd players. Let's say we have two components, A & B, which may or may not sound the same. Let's *not* say they are two cables.. let's say they are two amps with slightly different frequency response or distortion. Or they are two codecs. Or whatever, anything that we can all agree *may* or *may not* sound the same. First, in my experience, I can choose what to listen for, in listening to music. I can listen for the sounds of instruments, the quality of the dance invoked by the rhythm, the feelings of the music, the sound of distortion, the treble, the bass, and so on. Let's say A & B do sound different, and the difference between A & B is in the bass. But during the test I only listen to the treble. So we have the null result. Would you not agree this is a misleading result? Mike |
#2
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
|
#4
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
|
#6
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
|
#7
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
wrote in message
... First, in my experience, I can choose what to listen for, in listening to music. I can listen for the sounds of instruments, the quality of the dance invoked by the rhythm, the feelings of the music, the sound of distortion, the treble, the bass, and so on. Let's say A & B do sound different, and the difference between A & B is in the bass. But during the test I only listen to the treble. So we have the null result. Would you not agree this is a misleading result? All experiments produce some erroneous results. That's why they are usually done repeatedly with different people. |
#8
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
|
#9
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
On 23 Feb 2006 00:31:29 GMT, "bob" wrote:
A negative ABX test means only one thing: that the subject could not reliably distinguish between the two sounds in that test. Forgive the pedantry, but I don't think it actually means even that much. I would say it means that the test has failed to provide sufficient evidence to conclude that the subject could reliably distinguish between the two sounds in that test. Do sufficient identical tests with the same result and we might well be able to justify your stronger conclusion. |
#11
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
Stewart Pinkerton wrote:
On 22 Feb 2006 00:50:49 GMT, wrote: Let's say we have two components, A & B, which may or may not sound the same. Let's *not* say they are two cables.. let's say they are two amps with slightly different frequency response or distortion. Or they are two codecs. Or whatever, anything that we can all agree *may* or *may not* sound the same. Fine. First, in my experience, I can choose what to listen for, in listening to music. I can listen for the sounds of instruments, the quality of the dance invoked by the rhythm, the feelings of the music, the sound of distortion, the treble, the bass, and so on. Fine. Let's say A & B do sound different, and the difference between A & B is in the bass. But during the test I only listen to the treble. So we have the null result. Would you not agree this is a misleading result? No, I'd say that you ran the test incompetently, due to not using full-spectrum music for the comparison. I have a whole selection of favourite tracks that I use for listening to different aspects of sound when comparing audio gear. All the way from organ music and heavy drumming, to delicate cymbal brushwork, to single female voice to choral to full orchestra to jazz ensemble. Even pink noise and clicks for a *really* sensitive comparison. -- Stewart Pinkerton | Music is Art - Audio is Engineering Okay, thanks everyone for the explanation. I guess you would say, Stewart, that audio scientists are pretty well aware of what music to use and how to train their listeners. In other words, the possible difficult with ABX testing---namely, the possibility that subjects don't listen to the right things---is not, in practice, a major problem. And null results are nothing more than null results, perhaps *suggesting* that two compoments A & B are difficult to distinguish under those conditions, but not eliminating that anyone can hear them. With regard to cables, we have so many null results, and such an bovious absense of positive results, that one can be confident about their inaudibility. Is this all how you would state it? Mike |
#12
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
bob wrote:
wrote: Let's say A & B do sound different, and the difference between A & B is in the bass. But during the test I only listen to the treble. So we have the null result. Would you not agree this is a misleading result? No. I would only agree that you have misinterpreted the result. A negative ABX test means only one thing: that the subject could not reliably distinguish between the two sounds in that test. We cannot assume that the subject would always fail to distinguish those two sounds. And we cannot assume that other subjects would fail to distinguish those two sounds. Negative results don't tell us much. Positive results tell us more. A positive result (verified, which in practice probably means two positive results) tells us that the two sounds really are distinguishable. (That's why we keep asking, where are the positive results? It is their absence that is really telling.) Now, let's go back to your example, and ask a question: Why was the subject only paying attention to the treble? Could it be that, during his open listening, he thought he heard a distinct difference in the treble? If so, then this negative test does tell us something: It tells us that the difference he thought he heard may be illusory. We could make your test better by providing some training to our subject. Since we know that the amps differ in bass response, we could use an equalizer to exaggerate that difference, and let the subject practice on that first. This would help him focus on the real difference. Okay, thanks for the explanation. Now, on to one more question. My own experience is that some qualities of an audio system need listening for some time to discern. Normally I would like to hear an entire phrase, at least, to evaluate how the articulation is presented and whether that makes musical sense in relation to the phrase. I don't feel that I can be aware of qualities such as these in a 2-second snippet such as Arny uses. So what if two systems A and B do audibly differ in their presentation of these phrase-long qualities. What if we test them using ultra-short snippets, however. Is this a valid test? Under what conditions will it reveal an audible distinction between the systems? I know I may have phrased this in an unclear way, but go ahead and give me the explanation. Even though I responding to your post, bob, I invite Stewart and any other interested party to give me an explanation and/or clarify my language. Mike |
#13
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
|
#14
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
Ed Seedhouse wrote:
On 23 Feb 2006 00:31:29 GMT, "bob" wrote: A negative ABX test means only one thing: that the subject could not reliably distinguish between the two sounds in that test. Forgive the pedantry, but I don't think it actually means even that much. I would say it means that the test has failed to provide sufficient evidence to conclude that the subject could reliably distinguish between the two sounds in that test. Do sufficient identical tests with the same result and we might well be able to justify your stronger conclusion. His conclusion was that the subject couldn't reliably distinguish the two sounds in *that* test. It doesn't rule out that performance could be better on a subsequent test. -- -S "If men were angels, no government would be necessary." - James Madison (1788) |
#15
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
On 24 Feb 2006 00:40:24 GMT, wrote:
Stewart Pinkerton wrote: On 22 Feb 2006 00:50:49 GMT, wrote: Let's say we have two components, A & B, which may or may not sound the same. Let's *not* say they are two cables.. let's say they are two amps with slightly different frequency response or distortion. Or they are two codecs. Or whatever, anything that we can all agree *may* or *may not* sound the same. Fine. First, in my experience, I can choose what to listen for, in listening to music. I can listen for the sounds of instruments, the quality of the dance invoked by the rhythm, the feelings of the music, the sound of distortion, the treble, the bass, and so on. Fine. Let's say A & B do sound different, and the difference between A & B is in the bass. But during the test I only listen to the treble. So we have the null result. Would you not agree this is a misleading result? No, I'd say that you ran the test incompetently, due to not using full-spectrum music for the comparison. I have a whole selection of favourite tracks that I use for listening to different aspects of sound when comparing audio gear. All the way from organ music and heavy drumming, to delicate cymbal brushwork, to single female voice to choral to full orchestra to jazz ensemble. Even pink noise and clicks for a *really* sensitive comparison. Okay, thanks everyone for the explanation. I guess you would say, Stewart, that audio scientists are pretty well aware of what music to use and how to train their listeners. In other words, the possible difficult with ABX testing---namely, the possibility that subjects don't listen to the right things---is not, in practice, a major problem. Yes. And null results are nothing more than null results, perhaps *suggesting* that two compoments A & B are difficult to distinguish under those conditions, but not eliminating that anyone can hear them. That's true, although the available evidence suggests that those who do claim superior hearing acuity, invarably crash and burn when put to the test. This does suggest that there are no real 'Golden Ears', and that differences among cables in particular are simply *not* audible to any human. If you want the best possible chance, use children brought up in rural areas. They'll have the highest sensitivity and the widest frequency range. Self-important middle-aged hi-fi magazine reviewers and 'high end' dealers are just about the *worst* choice! :-) With regard to cables, we have so many null results, and such an bovious absense of positive results, that one can be confident about their inaudibility. Is this all how you would state it? That's pretty much it, and also applies to mainstream amplifiers and CD players. -- Stewart Pinkerton | Music is Art - Audio is Engineering |
#16
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
|
#17
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
Listening for a few minutes with highly repetitive material is best.
wrote: bob wrote: wrote: Let's say A & B do sound different, and the difference between A & B is in the bass. But during the test I only listen to the treble. So we have the null result. Would you not agree this is a misleading result? No. I would only agree that you have misinterpreted the result. A negative ABX test means only one thing: that the subject could not reliably distinguish between the two sounds in that test. We cannot assume that the subject would always fail to distinguish those two sounds. And we cannot assume that other subjects would fail to distinguish those two sounds. Negative results don't tell us much. Positive results tell us more. A positive result (verified, which in practice probably means two positive results) tells us that the two sounds really are distinguishable. (That's why we keep asking, where are the positive results? It is their absence that is really telling.) Now, let's go back to your example, and ask a question: Why was the subject only paying attention to the treble? Could it be that, during his open listening, he thought he heard a distinct difference in the treble? If so, then this negative test does tell us something: It tells us that the difference he thought he heard may be illusory. We could make your test better by providing some training to our subject. Since we know that the amps differ in bass response, we could use an equalizer to exaggerate that difference, and let the subject practice on that first. This would help him focus on the real difference. Okay, thanks for the explanation. Now, on to one more question. My own experience is that some qualities of an audio system need listening for some time to discern. Normally I would like to hear an entire phrase, at least, to evaluate how the articulation is presented and whether that makes musical sense in relation to the phrase. I don't feel that I can be aware of qualities such as these in a 2-second snippet such as Arny uses. So what if two systems A and B do audibly differ in their presentation of these phrase-long qualities. What if we test them using ultra-short snippets, however. Is this a valid test? Under what conditions will it reveal an audible distinction between the systems? I know I may have phrased this in an unclear way, but go ahead and give me the explanation. Even though I responding to your post, bob, I invite Stewart and any other interested party to give me an explanation and/or clarify my language. Mike |
#18
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
Stewart Pinkerton wrote:
On 24 Feb 2006 00:40:24 GMT, wrote: With regard to cables, we have so many null results, and such an bovious absense of positive results, that one can be confident about their inaudibility. Is this all how you would state it? That's pretty much it, and also applies to mainstream amplifiers and CD players. Except for the ones that don't. Some of the ones that don't are the result of incompetent design or defective manufacture. Others are the result of what I would call "competent but misguided design"--like cables with such extreme RLC values that they really do alter the signal enough to be heard. There are always provisos about when things sound the same, and when we leave them out we leave ourselves open to the "everything sounds the same" canard. bob |
#19
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
bob wrote:
wrote: My own experience is that some qualities of an audio system need listening for some time to discern. Normally I would like to hear an entire phrase, at least, to evaluate how the articulation is presented and whether that makes musical sense in relation to the phrase. I don't feel that I can be aware of qualities such as these in a 2-second snippet such as Arny uses. Note: You are assuming your conclusion here. I can see where you get that idea, but let me clarify. I'm saying that I'm aware of this "thing," this "quality" of the reproduction. Let's call it quality A. Quality A might be something like the relationship of the articulation to the phrasing. I can only become *aware*, *conscious*, of quality A, *in this described form*, by listening to a long phrase. Now, what I understand that you are telling me, is that whatever differences in sound contribute to this quality, can also be heard in much shorter phrases. I understand that this might be true, but what I'm suggesting is that I won't be aware of them in the same way. I simply *can't* be aware of the feeling of a phrase if I'm not hearing the whole phrase. However, I *can* be aware of differences in tonal balance, attack, detail, etc. in short snippets. So what I understand you are saying is that any difference which contributes to quality A, will be audible in some form in the short snippet. Is this how you would state it? So what if two systems A and B do audibly differ in their presentation of these phrase-long qualities. What if we test them using ultra-short snippets, however. Is this a valid test? Yes. Under what conditions will it reveal an audible distinction between the systems? Under all conditions (assuming the proper test protocols are followed). The problem is that I don't accept your assumption. I'm not assuming anything other than: I can't be aware of the quality of a phrase without hearing a whole phrase. I'm willing to accept that whatever signals contribute to that quality *can* be heard in snippets. However I have some more questions about this, but let's just start here and see if you agree with how I've phrased it. Neither, I suspect, would anyone in the psychoacoustics field. So far as we know (and we know plenty), there is no difference that you can hear in an entire phrase that cannot also be heard in some fraction of that phrase. If you doubt this, then it's up to you to demonstrate that some such difference actually exists. You're bucking the paradigm, here, and the burden's on you. bob |
#20
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
Stewart Pinkerton wrote:
On 24 Feb 2006 00:41:00 GMT, wrote: My own experience is that some qualities of an audio system need listening for some time to discern. Normally I would like to hear an entire phrase, at least, to evaluate how the articulation is presented and whether that makes musical sense in relation to the phrase. I don't feel that I can be aware of qualities such as these in a 2-second snippet such as Arny uses. So use a long-term test (shrug). Nothing in the rules to preclude it, it's simply that *decades* of exprience by people who do this for a living (the R&D guys at Revel, B&W etc) suggests that long-term listeing is *less* sensitive to small differences. If you can prove this to be incorrect, then good for you. So what if two systems A and B do audibly differ in their presentation of these phrase-long qualities. What if we test them using ultra-short snippets, however. Is this a valid test? Under what conditions will it reveal an audible distinction between the systems? It's a valid test, and the weight of available evidence suggests that your 'experience' is incorrect. The way to test this is to do a long-term comparison - but it must still be level-matched and blind. As I described more extensively in my post to Bob, my experience is that certain qualities of whole phrases can become conscious, in that specific form, only from listening to a specific phrase. In that sense my "experience" is not incorrect, it's just describing my experience. I'll ask you the same thing I'm asking him: Are you saying that any audible difference which I might describe in phrase-long terms, translate into an audible difference in "snippets".. which I would describe, or become conscious of, in some other form? In other words, listening to a whole phrase, I might say I'm aware of "phrasing", but listening to a snippet, I would describe the same difference as "attack"? Is this approximately the right idea? Mike |
#21
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
wrote:
bob wrote: wrote: My own experience is that some qualities of an audio system need listening for some time to discern. Normally I would like to hear an entire phrase, at least, to evaluate how the articulation is presented and whether that makes musical sense in relation to the phrase. I don't feel that I can be aware of qualities such as these in a 2-second snippet such as Arny uses. Note: You are assuming your conclusion here. I can see where you get that idea, but let me clarify. I'm saying that I'm aware of this "thing," this "quality" of the reproduction. Let's call it quality A. Quality A might be something like the relationship of the articulation to the phrasing. I can only become *aware*, *conscious*, of quality A, *in this described form*, by listening to a long phrase. Now, what I understand that you are telling me, is that whatever differences in sound contribute to this quality, can also be heard in much shorter phrases. I understand that this might be true, but what I'm suggesting is that I won't be aware of them in the same way. But if all you're trying to do is answer the question, "Can I hear a difference?" then it doesn't matter that you won't be aware of them in the same way. And if you're trying to answer a question other than "Can I hear a difference?"--like, for example, "What difference do I hear?"--then a DBT in which you listen only to very short snippets of sound may not be the right tool for the job. I simply *can't* be aware of the feeling of a phrase if I'm not hearing the whole phrase. However, I *can* be aware of differences in tonal balance, attack, detail, etc. in short snippets. So what I understand you are saying is that any difference which contributes to quality A, will be audible in some form in the short snippet. Is this how you would state it? Yeah, because at bottom, whatever you're hearing when you listen to that whole phrase has to be the result of differences in level, frequency, and/or timing at particular moments in that phrase. Those are broad categories, but I think that covers everything, doesn't it? And level, frequency and timing differences are quite detectable in DBTs. bob |
#22
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
bob wrote:
wrote: bob wrote: wrote: My own experience is that some qualities of an audio system need listening for some time to discern. Normally I would like to hear an entire phrase, at least, to evaluate how the articulation is presented and whether that makes musical sense in relation to the phrase. I don't feel that I can be aware of qualities such as these in a 2-second snippet such as Arny uses. Note: You are assuming your conclusion here. I can see where you get that idea, but let me clarify. I'm saying that I'm aware of this "thing," this "quality" of the reproduction. Let's call it quality A. Quality A might be something like the relationship of the articulation to the phrasing. I can only become *aware*, *conscious*, of quality A, *in this described form*, by listening to a long phrase. Now, what I understand that you are telling me, is that whatever differences in sound contribute to this quality, can also be heard in much shorter phrases. I understand that this might be true, but what I'm suggesting is that I won't be aware of them in the same way. But if all you're trying to do is answer the question, "Can I hear a difference?" then it doesn't matter that you won't be aware of them in the same way. And if you're trying to answer a question other than "Can I hear a difference?"--like, for example, "What difference do I hear?"--then a DBT in which you listen only to very short snippets of sound may not be the right tool for the job. I simply *can't* be aware of the feeling of a phrase if I'm not hearing the whole phrase. However, I *can* be aware of differences in tonal balance, attack, detail, etc. in short snippets. So what I understand you are saying is that any difference which contributes to quality A, will be audible in some form in the short snippet. Is this how you would state it? Yeah, because at bottom, whatever you're hearing when you listen to that whole phrase has to be the result of differences in level, frequency, and/or timing at particular moments in that phrase. Those are broad categories, but I think that covers everything, doesn't it? And level, frequency and timing differences are quite detectable in DBTs. Okay, I follow you. Now let's say we have two snippets or test tones: A & B. They differ somewhat in level, frequency, and/or timing. If they differ a lot, then we know the difference can be audible. If they differ a miniscule amount (e.g. two cables), we have pretty strong evidence the difference is not audible. Somewhere in-between are the set of differences which may or may not be audible.. we have to actually run the test to find out, and we have to understand that test conditions and choice of subjects will affect the result. Let's say I am the subject in a quick-switch test between A & B. For the difference to be audible to me, several things have to happen: 1. The ear (physical structure and aural nerve) has to be sensitive enough to detect the difference. 2. The resolution of the memory store has to be good enough, and the internal comparison to memory accurate enough, to reveal the difference. 3. I have to become conscious of the difference. Would you not agree that all three things have to be true in order to hear a difference? Mike |
#23
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
On 25 Feb 2006 17:30:24 GMT, "bob" wrote:
Stewart Pinkerton wrote: On 24 Feb 2006 00:40:24 GMT, wrote: With regard to cables, we have so many null results, and such an bovious absense of positive results, that one can be confident about their inaudibility. Is this all how you would state it? That's pretty much it, and also applies to mainstream amplifiers and CD players. Except for the ones that don't. Some of the ones that don't are the result of incompetent design or defective manufacture. Others are the result of what I would call "competent but misguided design"--like cables with such extreme RLC values that they really do alter the signal enough to be heard. There are always provisos about when things sound the same, and when we leave them out we leave ourselves open to the "everything sounds the same" canard. That's why I specified 'mainstream', by which I mean properly engineered bit otherwise unexceptional items, exemplified by the products of say Rotel, Yamaha, Denon, Pioneer, Sony, Arcam and ASudiolab (among others too numerous to mention). I very specifically do *not* mean so-called 'high end' stuff 'designed' by self-acclaimed 'gurus' with little or no technical background, vide Julian Vereker, Roy Gandy and Peter Qvortrup. Let's not even go near George Cardas.... -- Stewart Pinkerton | Music is Art - Audio is Engineering |
#24
Posted to rec.audio.high-end
|
|||
|
|||
double-blind testing
Stewart Pinkerton wrote:
On 25 Feb 2006 17:38:08 GMT, wrote: Stewart Pinkerton wrote: On 24 Feb 2006 00:41:00 GMT, wrote: My own experience is that some qualities of an audio system need listening for some time to discern. Normally I would like to hear an entire phrase, at least, to evaluate how the articulation is presented and whether that makes musical sense in relation to the phrase. I don't feel that I can be aware of qualities such as these in a 2-second snippet such as Arny uses. So use a long-term test (shrug). Nothing in the rules to preclude it, it's simply that *decades* of exprience by people who do this for a living (the R&D guys at Revel, B&W etc) suggests that long-term listeing is *less* sensitive to small differences. If you can prove this to be incorrect, then good for you. So what if two systems A and B do audibly differ in their presentation of these phrase-long qualities. What if we test them using ultra-short snippets, however. Is this a valid test? Under what conditions will it reveal an audible distinction between the systems? It's a valid test, and the weight of available evidence suggests that your 'experience' is incorrect. The way to test this is to do a long-term comparison - but it must still be level-matched and blind. As I described more extensively in my post to Bob, my experience is that certain qualities of whole phrases can become conscious, in that specific form, only from listening to a specific phrase. In that sense my "experience" is not incorrect, it's just describing my experience. I'll ask you the same thing I'm asking him: Actually no, you're describing an anecdote, you provide no evidence that you can *reliably and repeatably* do this when you don't *know* what's playing. I think that you think I'm claiming I hear a *difference* between two things in this quality. I'm saying something much simpler.. listening to a single device, this is one facet of musical experience which can only become conscious by listening to a whole phrase. Are you saying that any audible difference which I might describe in phrase-long terms, translate into an audible difference in "snippets".. which I would describe, or become conscious of, in some other form? In other words, listening to a whole phrase, I might say I'm aware of "phrasing", but listening to a snippet, I would describe the same difference as "attack"? Is this approximately the right idea? The idea is that there's nothing that you can *reliably and repeatably* identify in a long listening session, which will not be equally or better discerned in a 'snippet' comparison. Right, I understand that's the assertion. See my post to Bob, but I have the same question for you: To be aware of a difference, does not all the following have to happen: 1. The ear/aural nerve must resolve the difference 2. The neural memory store of presentation A must have sufficient resolution, and the comparison to presentation B must allow sufficient resolution to identify the difference 3. I must become conscious of the information "computed" by the comparison function I'm sorry if my language is a bit ambigious here; I am not very familiar with the neural models of ear/brain functioning. I would like to go to the Caltech library again and get some books on these topics, but I would appreciate any explanation that you feel interested in giving. Mike |
Reply |
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Forum | |||
Yet another DBT post | High End Audio | |||
A comparative versus evaluative, double-blind vs. sighted control test | High End Audio | |||
science vs. pseudo-science | High End Audio | |||
Acoustically transparent but opaque material for blind speaker testing? | General | |||
Speaker cables | Pro Audio |