Hm, honestly, I couldn't hear a notable difference.
I tried both "intensive listening" and ABX [of aa/bb] on my O2/HD650 combo, as well as some quick listening on my stereo chain (Rotel RSP960AX/RB971 + ancient Visaton Atlas Compact 4 way cabinets); all fed from a optical connected no-name DAC.
Either the differences are minuscle enough to require both well trained ears and repeated, intensive listening, *or* - what I doubt - they are totally drowned out by my chain; in which case I suspect the DAC to be the culprit (should really make up my mind on what DAC to build instead).
Anyway, looking forward to the solution 🙂
I tried both "intensive listening" and ABX [of aa/bb] on my O2/HD650 combo, as well as some quick listening on my stereo chain (Rotel RSP960AX/RB971 + ancient Visaton Atlas Compact 4 way cabinets); all fed from a optical connected no-name DAC.
Either the differences are minuscle enough to require both well trained ears and repeated, intensive listening, *or* - what I doubt - they are totally drowned out by my chain; in which case I suspect the DAC to be the culprit (should really make up my mind on what DAC to build instead).
Anyway, looking forward to the solution 🙂
I used my ODAC, cheap tweaked too (better sound that the original, like my AV Marantz SR4500, iFi iCAN, speakers, headphones and...).
Last edited:
If anybody is trying to listen on Windows and not using ASIO drivers, to avoid sound quality degradation by Windows due to automatic sample rate conversion, it's necessary to go to the control console sound settings, and set the default sound format to 16/48, since that's the format of the test files used in this thread. See image below for reference which shows the default setting:
Attachments
I imported the wav files into Goldwave and subtracted all the combinations. Some were less diff than others but I never saw the "near silence" that OP posted.
The result sounds like they just all had slightly different gains, and maybe some distorted/were louder in the high frequency range, ala the symbols.
Perhaps there was still some offset in the timing of each tracks, or a mild difference in amplitudes. I can post all the diff files once I can figure out how. I tried to auto-gain set 2 files and then diff them but this didn't seem to work well, it produced odd differences, semi-random loud spots. Probably has to do with the algorithm to auto-gain in Goldwave. Perhaps with different settings you could manage to minimize the level differences and do a better diff.
The result sounds like they just all had slightly different gains, and maybe some distorted/were louder in the high frequency range, ala the symbols.
Perhaps there was still some offset in the timing of each tracks, or a mild difference in amplitudes. I can post all the diff files once I can figure out how. I tried to auto-gain set 2 files and then diff them but this didn't seem to work well, it produced odd differences, semi-random loud spots. Probably has to do with the algorithm to auto-gain in Goldwave. Perhaps with different settings you could manage to minimize the level differences and do a better diff.
The tracks are all rather distorted, probably in no small part due to the D-A-D chain. Whatever extra distortion the test opamps may be adding seems to be mostly swamped. That's the way it seems to me, anyway.
There are enough clicks and pops that it sure sounds like it came from a record, but maybe partially cleaned up. And, if from a record that might account for some of the distortion too.
However, the cymbal tails in bb sound every so slightly phasey (although they don't sound all that much like cymbals anymore, which makes trying to notice small deviations from a familiar type of smooth, clean, realistic cymbal sound pretty much a no-starter for me), but I think with some more practice I just might be able to use that very little phasey effect to ABX it against another track, although I'm not sure if the phasiness is due to the opamp, or the RC network at it's input, as is shown in the schematic.
Other than the phasey effect, any differences in HF graininess in cymbal tails are very small, if reliably distinguishable at all compared to all the other swampy grit and graininess in the tracks.
Ideally for a test like this it would seem like a superbly clean audio path, the very best converters available, and 24-bit tracks would help a whole lot to clearly hear whatever there may be to hear from the test circuits themselves.
In fact, just 24-bit tracks might help a lot, even if the source material was less than 24-bits. Whatever distortion may be added is probably largely down in the low bits we don't have. And once that distortion is added, we now have a signal with more effective bit-depth that could be captured at better resolution.
In the meantime, though, I'll keep practicing my listening.
There are enough clicks and pops that it sure sounds like it came from a record, but maybe partially cleaned up. And, if from a record that might account for some of the distortion too.
However, the cymbal tails in bb sound every so slightly phasey (although they don't sound all that much like cymbals anymore, which makes trying to notice small deviations from a familiar type of smooth, clean, realistic cymbal sound pretty much a no-starter for me), but I think with some more practice I just might be able to use that very little phasey effect to ABX it against another track, although I'm not sure if the phasiness is due to the opamp, or the RC network at it's input, as is shown in the schematic.
Other than the phasey effect, any differences in HF graininess in cymbal tails are very small, if reliably distinguishable at all compared to all the other swampy grit and graininess in the tracks.
Ideally for a test like this it would seem like a superbly clean audio path, the very best converters available, and 24-bit tracks would help a whole lot to clearly hear whatever there may be to hear from the test circuits themselves.
In fact, just 24-bit tracks might help a lot, even if the source material was less than 24-bits. Whatever distortion may be added is probably largely down in the low bits we don't have. And once that distortion is added, we now have a signal with more effective bit-depth that could be captured at better resolution.
In the meantime, though, I'll keep practicing my listening.
Last edited:
You may not use direct file subtraction, because of time differences in us between the files. The only way is to use Audio Diffmaker that eliminates phase differences. This is the very basic. Then you get just noise as a result of subtraction. Direct subtraction shows linear issues, not distortion, which is nonlinear. Distortion in the chain used is well below 0.001%.
I was afraid that someone would come with direct subtraction. That is why I refused to post the original file. Same mistakes were done long ago in analog subtraction - Hafler.
Seems you see insulting is ok, so long as it is only going in one direction.
When it's reciprocated you have an issue with it.
I call that hypocrisy.
You're the one making extraordinary claims and then refusing to give a shred of evidence to back it up. Further, when called out on it, you then imply that people who don't believe your completely unsubstantiated assertions without evidence are idiots.
How are the claims made in your post relevant to the quote of my post which you have included?
They are not. You are conflating different things and making implications on those conflations that you invented that there was no evidence whatsoever to construe an implied conflation.
Last edited:
I'm going to have to read this DiffMaker page and meditate on your post for a bit. 🙂
My first reaction is: " Audio Diffmaker ... eliminates phase differences" ... aren't phase differences, differences? By phase differences, you mean slightly skewed start times? And/or slightly skewed amplitudes? As long as Diffmaker first time-aligns the data (bit right/left shifting, not compressing/expanding time altering throughout the wave file, which I'm thinking is cheating) and then scales one or both waveforms to a common amplitude, and/or calculates a scalar between them and grades the diff based on how close to that scalar it gets... then I agree with Diffmaker.
Anyway, I'll go study.
My first reaction is: " Audio Diffmaker ... eliminates phase differences" ... aren't phase differences, differences? By phase differences, you mean slightly skewed start times? And/or slightly skewed amplitudes? As long as Diffmaker first time-aligns the data (bit right/left shifting, not compressing/expanding time altering throughout the wave file, which I'm thinking is cheating) and then scales one or both waveforms to a common amplitude, and/or calculates a scalar between them and grades the diff based on how close to that scalar it gets... then I agree with Diffmaker.
Anyway, I'll go study.
BTW... In this case I think the differences are completely academic/coincidence. If you recorded the same source data on the same op amp M/N and diff S/Ns, each 10x in a row I bet you would see a distribution of values and could compute a mean/std deviation, etc.
Another interesting comparison would be (IMO) a low distortion [Op] Amp vs. say a very expensive hi-fi amp (perhaps with much higher THD), or better yet, various other components in the chain more likely to mess up the signal.
-BK
IN theory DiffMaker is a great idea. It's a method quite often used in test and measurement.
In reality there are so many variables to take into account it almost brings one back to square one, unless those differences are huge. If they are down on the noise floor things are going to get difficult indeed.
Also, it's not always true AB either, because as soon as you introduce compensation for different cables for example, you throw a whole lot of other variables into the mix.
Even if you do identify a change caused by the difference, understanding the mechanism is not revealed by Diffmaker. It may show you what, but it won't show you why. That's where knowledge and experience help.
Lastly is it more sensitive or revealing than other test equipment out there? If it is not, you are back with relying on your ears mainly.
In reality there are so many variables to take into account it almost brings one back to square one, unless those differences are huge. If they are down on the noise floor things are going to get difficult indeed.
Also, it's not always true AB either, because as soon as you introduce compensation for different cables for example, you throw a whole lot of other variables into the mix.
Even if you do identify a change caused by the difference, understanding the mechanism is not revealed by Diffmaker. It may show you what, but it won't show you why. That's where knowledge and experience help.
Lastly is it more sensitive or revealing than other test equipment out there? If it is not, you are back with relying on your ears mainly.
Last edited:
There were some speculations about distortion originated in the D/A ==> A/D chain in this test. Therefore, I am attaching basic measurements of distortion in the chain.
Second, I have created 4 difference files in Audio Diffmaker, they cover all 4 opamps used. The files can be downloaded from
http://pmacura.cz/optestdiffiles.zip
Once again I would like to repeat that direct subtraction is a poor idea, because of small time/phase shifts that get magnified as frequency is increased. Please take into account that you do not listen to direct sample values, but the signal created by the reconstruction filter. Thus, information of sample amplitude difference is meaningless.
Second, I have created 4 difference files in Audio Diffmaker, they cover all 4 opamps used. The files can be downloaded from
http://pmacura.cz/optestdiffiles.zip
Once again I would like to repeat that direct subtraction is a poor idea, because of small time/phase shifts that get magnified as frequency is increased. Please take into account that you do not listen to direct sample values, but the signal created by the reconstruction filter. Thus, information of sample amplitude difference is meaningless.
Attachments
For this reason, I always have my doubts when audiophiles make a choice for R.L. Jones. To my ears, there always is a kind of graininess to her voice. Her singing goes through some echo device, which will not only distort, but may also introduce comb filtering.
It is not easy to make a choice of a music sample. When I tried Beethoven, there were people who did not like it, and I might continue through all genres. I thought that this one might have been acceptable for most people. The recording (original) has audible distortion, but which recording has not, except for well recorded classical music. If we start to analyze with headphones, piece by piece of any recording, we almost always find some kind of flaws.
I can see only 3 attached image files apart from the zip diff files. Two spectrum with test tones and one THD. Was there a there a spectrum control with no test tones taken? That is, a spectrum of the noise floor looking only at the testing chain internally generated spectral components?
- Status
- Not open for further replies.