Sorry for the confusion that I caused, but after having checked everything a 1000 times, this should be the correct 4+4 test.
Dropbox - NOS-44.1 - Simplify your life
Succes and I'm looking forward to receive your PM's to me.
Hans
Dropbox - NOS-44.1 - Simplify your life
Succes and I'm looking forward to receive your PM's to me.
Hans
FYI - Below, is a little bit of information from the creator of PGGB, regarding how Hans' test files were resampled by his conversion software. With a few edits by me.
_______________________________________________________________________________
The upsampled files use my own proprietary dithering, which I use for lower (sample) rates. It also reduces aliased compnents in CD audio. The (filter) taps are dependent on file length, but I think I saw lengths between 18- 52 Million (tap lengths for resampling Hans' test files). The Windowed Sinc is adaptive, and it has heuristics that will choose the windowed parameters adaptively for a given track length to get the right balance between maximizing reconstruction accuracy and maximizing stop band attenuation.
_______________________________________________________________________________
The upsampled files use my own proprietary dithering, which I use for lower (sample) rates. It also reduces aliased compnents in CD audio. The (filter) taps are dependent on file length, but I think I saw lengths between 18- 52 Million (tap lengths for resampling Hans' test files). The Windowed Sinc is adaptive, and it has heuristics that will choose the windowed parameters adaptively for a given track length to get the right balance between maximizing reconstruction accuracy and maximizing stop band attenuation.
Last edited:
Clarification on the digital filter experiment experience
Our initial thought was that, a perfectly transparent resampling interpolation-filter would leave the matching track pairs sounding identical. While that is largely true, you may also notice that one track still sounds somewhat 'better' than the other. Dithering/noise-shaping of the least significant bits on the resampled tracks may be responsible for that, however, it also possible that should a particular resampling interpolation-filter more accurately reconstruct the signal, it then it reveals a 'better' sound in doing so.
You may also notice the above difference in sound, even when playing the tracks via a typical OS DAC with a chip-based interpolation-filter. The reason may be that, via an OS DAC the resampled test tracks are, effectively, passed through a cascade of two digital filters in series. One superior performing filter (the resampler's) followed by a second, inferior performing filter (the OS DAC's). The superior performing filter will dominate the overall performance of the two, especially, the transition-band slope and the stop-band frequency and suppression.
Therefore, those of you having only an OS DAC may also find this experiment enlightening. So, if you do perform it utilizing an OS DAC, please clearly state that fact in your listening results report to Hans. Thank, you.
Our initial thought was that, a perfectly transparent resampling interpolation-filter would leave the matching track pairs sounding identical. While that is largely true, you may also notice that one track still sounds somewhat 'better' than the other. Dithering/noise-shaping of the least significant bits on the resampled tracks may be responsible for that, however, it also possible that should a particular resampling interpolation-filter more accurately reconstruct the signal, it then it reveals a 'better' sound in doing so.
You may also notice the above difference in sound, even when playing the tracks via a typical OS DAC with a chip-based interpolation-filter. The reason may be that, via an OS DAC the resampled test tracks are, effectively, passed through a cascade of two digital filters in series. One superior performing filter (the resampler's) followed by a second, inferior performing filter (the OS DAC's). The superior performing filter will dominate the overall performance of the two, especially, the transition-band slope and the stop-band frequency and suppression.
Therefore, those of you having only an OS DAC may also find this experiment enlightening. So, if you do perform it utilizing an OS DAC, please clearly state that fact in your listening results report to Hans. Thank, you.
Hoi Hans, what conversions/recordings are we listening here?
Analog sampling of a NOS DAC at 44100Hz v.s. Up/down sampled via audacity?
I am a bit lost what this test wants to test.
If it is NOS v.s. oversampled, shouldnt there be recordings of both DACs at a very high sample frequency, like >-192KHz and compare these?
@ MVS0,
The previous test was: up-sample to 192Khz and downsample again to 44.1/16 all done with Audacity. First impression was that this lead to auditory differences.
This time a much more sophisticated FIR filter is used by PGGB with a length of over 1Mio coefficients, upsampled to 88.2 and downsampled again to 44.1/16.
Since we are testing whether processing the signal may result in audible differences, that’s why the test is currently restricted to 4+4 .wav files.
If any differences can be detected, it could give a clue why NOS and OS have a different sound profile.
Hans
The previous test was: up-sample to 192Khz and downsample again to 44.1/16 all done with Audacity. First impression was that this lead to auditory differences.
This time a much more sophisticated FIR filter is used by PGGB with a length of over 1Mio coefficients, upsampled to 88.2 and downsampled again to 44.1/16.
Since we are testing whether processing the signal may result in audible differences, that’s why the test is currently restricted to 4+4 .wav files.
If any differences can be detected, it could give a clue why NOS and OS have a different sound profile.
Hans
Looked at the first set of files "God give me strength" and see +/- 0.35dB level difference....
That's not surprising. The resampled tracks were up-sampled to 88.2 @ >16bits using an FIR interpolation-filter of well over than 1 million taps/coefficients, then down sampled back to 44.1 @ 16bits using dither. So, the LSBs are not identical between the original and the resampled versions of a given track.
The level difference is also in the lower frequencies. So this looks like a small scaling problem of the FIR coeficients. I will take closer look tomorrow. A real 0.35dB level diffence could be audible by itself.