I would say the conversion 'errors' have a much lower audible significance, in this instance. Personally, I find all DAC technology goes through a quality cycling over a time frame, in the daily use of it, and most likely the cheaper the 'bits' the worse this aspect is - but I don't hear this being particularly prominent here.
From cold, DACs are pretty dead, lacklustre - no life or 'spirit' to them - the very things the vinyl people pour scorn on. Then, they slowly build up an edginess or harshness, shrill or ear piercing qualities. Finally, they settle down, stabilise, and deliver a balanced sound, full of detail and 'musicality' - the irksome thing is that this process has to be gone through every time there is a cold start.
High quality, expensive digital shouldn't have these problems, or at least much less of it. Effectively, these are 'conversion errors' - and, yes, they are part of the perception story ...
Sure the digital side of the dac is responsible for the serious warm up artifacts you hear? I hear nothing so severe from my PCM1794A based dac which has passive I/V conversion and vacuum tube amplification. My suspicion has long been that the analog hardware (design and implementation) in many dacs is the culprit for much of the described ills of digital audio. Or it could just be that I have tin ears.. 
If this were the case, then I think you would see a lot of threads about "why does my XYZ player sound different every time I listen to it?"
My guess is that these differences are random in nature, making it pretty much impossible to pick up a change consistently. The change (if any) that a different input cap imparts on the sound should be a constant. ie it should have a particular sonic impact which is the same each and every time the playback / recording is done (this is what is being tested for here).
There of course could be exceptions like if for instance the cap was a huge one with long leads, and may be picking up RF (I guess that is a possibility with this test also due to the stringing together of the caps) or if Mooly had used a large X7R ceramic and placed it close to the speaker.
Tony.
That is what we want to hear
I thought this was a listening test, why is everyone using scopes and abx via foobar ...
The scopes is human nature The ABX is I think necessary to give people the confidence to do the sorting. If I couldn't get a good result in the ABX I wouldn't be bothered sorting all the files. If I can get a good result in the ABX that gives me the confidence that I won't be wasting my time doing the sorting test
edit: I think it is worthwhile hiding the results of the ABX until you are finished. If you don't know you got one right or wrong until the end your brain doesn't over analyse the situation, in this vein, perhaps just doing the sorting without ABX would provide potenitally better results!
Tony.
The ABX is I think necessary to give people the confidence to do the sorting. If I couldn't get a good result in the ABX I wouldn't be bothered sorting all the files. If I can get a good result in the ABX that gives me the confidence that I won't be wasting my time doing the sorting test edit: I think it is worthwhile hiding the results of the ABX until you are finished. If you don't know you got one right or wrong until the end your brain doesn't over analyse the situation, in this vein, perhaps just doing the sorting without ABX would provide potenitally better results!
Tony.
Last edited:
Thanks
I wasn't sure at first that getting them so identical was easily do able, but with care and using Audacity with the "trace" magnified to set the start point visually by eye it all fell into place.Conversion errors and how that possibly effects things is an unknown to me at an absolute level. I'm not qualified and haven't the specific knowledge to say on that.
What my thoughts are though is that the cap substitution is purely in the analogue domain, and we know the conviction of those that promote use of high end parts and the benefit (they feel) that those parts bring. So that difference if it exists is a constant running through the whole test. My feeling is that the digital side of the process, me making the files and the (digital) replay equipment you all use, although it may well have "random" differences each time you play back (and each time I record)... that these differences are audibly minute and in any case "a different kind" of difference to the properties of the parts under test.
Similar thoughts I suspect from Tony. And a good point mentioning why we don't see more of the "why does my XYZ player sound different every time I listen to it?" Presumably it is a non problem because any effect if there at all is below any audibility threshold. I would think the same argument applies for those making recordings too.
The test conditions for the two caps were identical and the test done in quiet conditions so there are no microphonic problems to worry over.
I thought this was a listening test, why is everyone using scopes and abx via foobar ...
Because they do
This is a good question. Rational approach is needed again and measurements and analysis is necessary. If you record the same music sample from CD player's analog output several times, make difference files and get consistent almost zero result, then the hardware used should be OK and you would not ask why it always sounds different.
In case of Mooly's files, even two files consequently recorded from the same setup do produce difference that is both measurable and audible. And this is, at least to me, a BIG problem, though there is a lot of effort to prepare a great test and this effort must be appreciated. I strongly suspect the TDA1540, the chip that does not use proper output anti-aliasing filter with sharp cut off of everything above Fs/2. Instead, it produces teeth-like output, which seems to be never twice same. This should be checked. Otherwise, if Mooly used different hardware, the test might have been great.
I'm pretty sure that in a taste test I'd be able to consistently tell the difference between apricots and plums even though not all apricots are exactly the same shape. Heck, I doubt I'd be thrown even if some of the plums were closer in size, shape and color to some of the apricots than they were to the other plums.Because they do
OTOH, after reading this thread, I get the idea that some folks couldn't tell the difference between a meatball and a potato without using a tape-measure.
Hi All I am talking about the analogue to digital conversion process that Karl used when recording the streams not the digital to analog conversion process, this is pretty well established and has few errors, but there does not exist an ADC anywhere that can sample an analog stream at a high sample rate and produce zero errors, some time you can get pretty large errors. So I fear to say no-one understood what I was trying to say and I will say again is there a big enough difference in sampling while recording that could mask the difference in capacitors. In other words there are two variables in each of the recordings.
Last edited:
This is much more like comparing apples of the same variety from different trees, or a wine maker tasting the grapes on different days and different parts of the vineyard while determining when to pick. In this case a refractometer comes in handy for measuring sugar content.
Thats a very valid opinion, one I am unable to prove or disprove with what I have available.
I'm just thinking aloud now... you record (sample) a track as I have done and then repeat the exercise. This "difference" we talk of appears when one is subtracted from the other. Again thinking aloud.. is subtracting files a valid approach to real audible differences or, is it as if for example, that you move all the books on shelf left or right a little. There is a "linear" difference but the basic info is the same on the shelf (think time scale) as it was before and as it becomes.
Is that understandable
We think of total accuracy in digital systems but what about long term drift of system clocks... irrelevant in a real sense yet perhaps very relevant when you have two side by side processes, the player, and the ADC in the PC. Could that be where differences show in file analysis ?
Mooly,
I've posted in your other listening test threads about track alignment errors. Same applies here. Your alignments are sloppy. Cool Edit allows maintaining cursor position when switching between track, and provides far better zoom control. Waveform display is interpolated as well, providing much more accurate waveform depiction compared to Audacity's crappy connect the dot display.
Loading the 14 tracks, B was found to have the earliest start time. Switching back and forth between B and other tracks a series of cue markers were made indicating alignment errors to single sample accuracy:
In above it is seen that start times vary by as much as 66 samples, roughly 1.5ms. This could impact how some sort files with ABX.
Alignment to single sample is not good enough for applying sample to sample difference (file difference) approach in analysis. This requires sampling files to higher rate, such as fs 192kHz and making further adjustment. Then amplitudes of waveforms need to be matched, preferably to levels better than 0.001dB. Once this is accomplished, difference results such as Pavel's are possible:
In above overview of tracks are shown. Yellow vertical line shows point chosen for making alignment and creating spectrum view.
With tracks aligned at this point, examination near end reveals alignment drift, this may be due to number of causes. Drift in clock frequency between recordings, sample dropout, and thermal effects as example possibilities. This is seen in difference track as increasing amplitude.
Pavel pointed out glitch in file difference spectrum around 4kHz. In his and above view closer examination reveals harmonic character, with spectral pattern recurring at about 8kHz. With closer study, this motif is also visible at 250Hz, 500Hz, 1kHz, 2kHz. This could be artifact of difference technique, but could also be subtle differences in group delay characteristic of the two test conditions, or even possibly harmonic distortion.
I've posted in your other listening test threads about track alignment errors. Same applies here. Your alignments are sloppy. Cool Edit allows maintaining cursor position when switching between track, and provides far better zoom control. Waveform display is interpolated as well, providing much more accurate waveform depiction compared to Audacity's crappy connect the dot display.
Loading the 14 tracks, B was found to have the earliest start time. Switching back and forth between B and other tracks a series of cue markers were made indicating alignment errors to single sample accuracy:
In above it is seen that start times vary by as much as 66 samples, roughly 1.5ms. This could impact how some sort files with ABX.
Alignment to single sample is not good enough for applying sample to sample difference (file difference) approach in analysis. This requires sampling files to higher rate, such as fs 192kHz and making further adjustment. Then amplitudes of waveforms need to be matched, preferably to levels better than 0.001dB. Once this is accomplished, difference results such as Pavel's are possible:
In above overview of tracks are shown. Yellow vertical line shows point chosen for making alignment and creating spectrum view.
With tracks aligned at this point, examination near end reveals alignment drift, this may be due to number of causes. Drift in clock frequency between recordings, sample dropout, and thermal effects as example possibilities. This is seen in difference track as increasing amplitude.
Pavel pointed out glitch in file difference spectrum around 4kHz. In his and above view closer examination reveals harmonic character, with spectral pattern recurring at about 8kHz. With closer study, this motif is also visible at 250Hz, 500Hz, 1kHz, 2kHz. This could be artifact of difference technique, but could also be subtle differences in group delay characteristic of the two test conditions, or even possibly harmonic distortion.
You don't need track alignment in a sorting test. The tracks don't even have to be the same length.
Just because some people are trying to avoid the controls by recasting this as an ABX test doesn't mean that it is one.
Mooly your recording ability just went out the window if that is what is under scrutiny.
Next time you think of such a tests ask for volunteers and send them each a hand full of capacitors, let them figure it out from there.
Can you hear a difference and if so which tracks sound the same? Better? Whats so difficult ...?
I might even try this, though I have the usual problem. The example at hand is outside the context of actual use, that is who would use a series connection of 5 cheap caps anywhere in a circuit. I have the same problem with some of Bateman's results, they lead some to make sweeping generalizations about the "sound" of passives when it's just that good engineering practice is left out of the equation.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.