Hard to see the point of doing anything like that. dw-1 and dw-2 are so muddied up as compared to the original wav file played through v03 on my system such that they no longer have much to do with how the FPGA versions sound. And, yes, dw-1 and dw-2 sound different from each other anyway. But the sound and the difference have nothing to do what what v03 and v04 sound like in reality. Its like you watch two different superman movies on a large screen, then someone wants to see if you can statistically tell the two movies apart by looking at comic book pictures of each movie. It makes not sense at all.
May I ask if anyone else here has listened to the files, and if they know what I'm talking about?
May I ask if anyone else here has listened to the files, and if they know what I'm talking about?
The chart in post #3122 displays the spectras from unmatched waveforms. I don't know how they are calculated and why there is such a difference. Considering this is freeware it is quite possible that it is caused by a bug in SW. Anyhow since the whole point of DeltaWave is about matching why put so much emphasis on that particular chart?
Closer look with Audacity shows that there is a delay of about 0.25s on PCM2DSD recordings. So comparing original spectras without removing the delay will quite likely result in different looking spectras.
Closer look with Audacity shows that there is a delay of about 0.25s on PCM2DSD recordings. So comparing original spectras without removing the delay will quite likely result in different looking spectras.
Last edited:
Have you matched the levels between the original and PCM2DSD recordings? Without matching the levels of course lower level recording sounds muddied up.
I added 2dB gain to both recordings which should make the levels more or less the same as in original. Here is a new download link:
https://www.dropbox.com/t/6NsDD8m7Ir4fVVaQ
Last edited:
I accidentally had 16-bit setting in Audacity for the +2dB files. They should work as well but here are +2dB 32-bit files which are similar to my previous recordings:
https://www.dropbox.com/t/tYIjsRZZj8TojQHk
https://www.dropbox.com/t/tYIjsRZZj8TojQHk
Because we don't understand it.
Could it be that the plot shows DFTs of a part of the whole recording and that due to a delay, different parts of the music are used for the different plots?
As I'm not the developer of DeltaWave so I don't know either but I still don't see that issue relevant for this discussion.
Rather than looking at that spectra, please listen to the recordings in my previous post and to Markw4's original in post #3114.
DeltaWave calculates a gain difference of about 0.1dB with the +2dB files compared to the original. That should be sufficiently well matched level for direct ABX.
As I did not remove the delay from PCM2DSD recordings that may be a giveaway if somebody does ABX against the original and has "Keep playback position when changing track" checked. IMO using such "tells" is akin to cheating. ABX between PCM2DSD recordings does not have such "tells".
Here are new PCM2DSD recordings with delay removed so less "tells" even against the original:
https://www.dropbox.com/t/sA9BIbNwsWL5lAuw
Here are new PCM2DSD recordings with delay removed so less "tells" even against the original:
https://www.dropbox.com/t/sA9BIbNwsWL5lAuw
Last edited:
You are right. Attached below is my playlist. I thought I clicked on entry 19 for comparison, but it must have been entry 21 or 22. No wonder it seemed a little worse than my long term memory of it.
EDIT: However, I will say that the imaging was better on the first file today, although there was a little bit of phasiness I thought odd. Sometimes that happens if there is a reflection behind the speakers, other times its because of the recording. Today I assumed I might have to check behind the speakers later.
EDIT: However, I will say that the imaging was better on the first file today, although there was a little bit of phasiness I thought odd. Sometimes that happens if there is a reflection behind the speakers, other times its because of the recording. Today I assumed I might have to check behind the speakers later.
Last edited:
Sorry, I don't buy your explanation unless you provide Foobar ABX scores. The difference especially between PCM2DSD recordings is very small. And as I said before it would be more useful if you can pinpoint the locations where you perceive a difference so that others can have a go as well.
But if you now feel that also "St. Louis Blues - 60s-dw2_+2dB_32_nodelay" sounds like the original recording, with the imaging still intact, then no need for Foobar ABX.
BTW this is a very good example of the importance of level matching in listening evaluation. Practically same recording with -2dB level sounds like a muddy, dynamic compressed copy, with lost imaging.
But if you now feel that also "St. Louis Blues - 60s-dw2_+2dB_32_nodelay" sounds like the original recording, with the imaging still intact, then no need for Foobar ABX.
BTW this is a very good example of the importance of level matching in listening evaluation. Practically same recording with -2dB level sounds like a muddy, dynamic compressed copy, with lost imaging.
Last edited:
I don't care what you buy. You bought into DeltaWave without a clue.
Regarding "St. Louis Blues - 60s-dw2_+2dB_32_nodelay", it sounds about like it did yesterday. When I went back to see what happened I did listen to the original wave file today, so my short term memory was now is use. Nothing has changed from yesterday.
Sorry, neither of your copies sound like the original. It would be foolish to believe they could. Mastering engineers stopped going to analog and then back to digital for file format conversion decades ago. It never sounds as clean and accurate as format conversion in the digital domain.
If you want to do fair tests you need to loopback the digital output of PCM2DSD and record it into a DSD file. Then use high quality professional software such as Pyramix to convert to PCM. Since that is apparently beyond your competence you substitute an excessively flawed analog process. No one should expect the resulting PCM files to sound accurate.
Regarding "St. Louis Blues - 60s-dw2_+2dB_32_nodelay", it sounds about like it did yesterday. When I went back to see what happened I did listen to the original wave file today, so my short term memory was now is use. Nothing has changed from yesterday.
Sorry, neither of your copies sound like the original. It would be foolish to believe they could. Mastering engineers stopped going to analog and then back to digital for file format conversion decades ago. It never sounds as clean and accurate as format conversion in the digital domain.
If you want to do fair tests you need to loopback the digital output of PCM2DSD and record it into a DSD file. Then use high quality professional software such as Pyramix to convert to PCM. Since that is apparently beyond your competence you substitute an excessively flawed analog process. No one should expect the resulting PCM files to sound accurate.
Last edited:
In other news, the Marcel dac with IanCanada clock has been running all night. Its starting to sound a lot better with some time.
However, a small problem cropped up. The reclocking board just before the Marcel dac aligns the clock with the data exactly in the middle of the data pulses. However, sometime during the night, it looks like I2SoverUSB slipped one 45/49MHz clock edge out of alignment with the divided 22/24MHz clock outputs. That put the BCK signal edge exactly aligned with data pulse edges just before going into Marcel's dac. Probably I need to introduce a small delay into Marcels dac so that it will work with either timing of the reclocked I2SoverUSB board. Maybe just a couple of inverters on the clock line to delay it slightly or something like that. Have to think about it a little more.
However, a small problem cropped up. The reclocking board just before the Marcel dac aligns the clock with the data exactly in the middle of the data pulses. However, sometime during the night, it looks like I2SoverUSB slipped one 45/49MHz clock edge out of alignment with the divided 22/24MHz clock outputs. That put the BCK signal edge exactly aligned with data pulse edges just before going into Marcel's dac. Probably I need to introduce a small delay into Marcels dac so that it will work with either timing of the reclocked I2SoverUSB board. Maybe just a couple of inverters on the clock line to delay it slightly or something like that. Have to think about it a little more.
Last edited:
Foobar ABX is essentially a software implementation of the test promoted by the ABX company in Troy, Michigan.
Over the years it has seen extensive criticism of the method, for acting as a strong randomising factor in Audio tests of small differences in audible fidelity.
In fact, the applied statistics combined with the very low number of data points, gives the ABX test promoted by the ABX company in Troy, Michigan very high statistical significance IF the null hypothesis is rejected, but very low statistical power.
For failure to reject the null hypothesis, it Lacs statistical significance and statistical power, in fact the test is extremely heavily weighted towards "avoiding false positives" that "false negatives" for subtle audible fidelity impairments become a near certainty.
A few years (might be by now decades) I did a statistical meta analysis of all available, published ABX tests, according to the method promoted by the ABX company in Troy, Michigan, that failed to reject the ",null hypothesis".
Meta analysis of multiple experiments is common in the medical field, I'm interested in this as I am no longer young and am suffering from a few chronic health conditions. Finding the most effective, cost effective and least side effect pre loaded medication is hence of great interest to me.
Back to audio, the new amalgamated "null result" data set was very large, giving high statistical power.
But the real clincher is that the larger result set of ONLY null results, suggested with a significance of better than .1 that in fact a reliable difference was perceived in the tests, but the actual statistics applied to this test were weighted excessively towards failing to reject the null hypothesis, as opposed to giving an equal weight to accepting or rejecting the null hypothesis.
This ultimately a result of using statistical analysis meant for large datasets to very sparse data sets.
Hence I recommend to regard the ABX tests, according to the method promoted by the ABX company in Troy, Michigan, as an utter waste of time, as their failure to reject the null hypothesis has neither statistical significance nor power.
Of course, it's your time.
There are enough documented methodologies, including by the ITU/EBU and the people who "invented" MP3 (talk to JJ - not sure where he hangs out these day's) that do not suffer the fundamental limitations of the ABX tests, according to the method promoted by the ABX company in Troy, Michigan.
I do not even want to go into Placebo/Nocebo, the fact that if a "blind" test is not blind to what is actually being tested and where the subject holds an opinion on the state of affairs is not really valid...
TU Dresden - WE HEAR WHAT WE EXPECT TO HEAR
Put simply, if we expect to hear no difference, we will not hear one, even if an actual difference, reliably audible to other subjects, is present. (Nocebo)
Equally, if we expect to hear a difference, we will hear one, even if there is no actual difference, and the fact that the two presentations are reliably audible to other subjects as identical. (Placebo)
Thor
Do you have a reliable, peer reviewed paper including a suitable test that offers both statistical significance and statistical power that support this assertion?
I find it rather cavalier and unscientific. Note, I am not saying you are wrong, but I'd like to see evidence in support of such general claims, when they are made.
Thor
Do you have any reliable, peer reviewed paper including a suitable test that offers both statistical significance and statistical power that does not support this assertion?
No need for ad hominem attacks. This was not a test for proving your listening capabilities nor was it an attempt to claim that resulting PCM files sound accurate. You were the one making it into such. Unfortunately your listening analysis failed and the fact that you did not even attempt to match levels casts a shadow on all your listening evaluation claims. Past, present and future.