Any idea how these "same but different" masters were created?

Topic: Any idea how these "same but different" masters were created? (Read 1121 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.

Any idea how these "same but different" masters were created?

2025-02-02 19:27:49

If you are bored enough: These test files are offered for free download as WAVE, FLAC and as scaMQA. And at three different DSD rates: https://bluecoastmusic.com/free-downloads

You would expect that a FLAC and a WAVE at same resolution would be bit-identical, but they are not.
Indeed, upsampling all the PCM to 384/32 float and running a fb2k bitcompare, the typical peak difference is around the -20 to -40 dB mark - the MQA distortion doesn't make for closer to zero.
(This but for the 192/24 pair which seem to only differ at dither.)

I am not asking about what is audible. I am curious if there is any well-known workflow that makes this apparent lossy nonsense.

I am not sure how ffmpeg does DSD to PCM conversion, but those appear to differ even more. First reaction was, aha that is due to the stupid amount of HF noise this format introduces. But even doing resampling to 24 kHz, they turn out to differ at -16 to -17 dB below their peaks; that is differences to each other, not to the PCM.

Edit:
Of course this is music too. You can listen to more of his nice four-string treatment at https://soundcloud.com/derekjones-music . (Not the same Derek Jones as the late Falling In Reverse guitarist.)

Re: Any idea how these "same but different" masters were created?

Reply #1 – 2025-02-02 22:07:37

Quote from: Porcus on 2025-02-02 19:27:49

You would expect that a FLAC and a WAVE at same resolution would be bit-identical, but they are not.
Indeed, upsampling all the PCM to 384/32 float and running a fb2k bitcompare, the typical peak difference is around the -20 to -40 dB mark - the MQA distortion doesn't make for closer to zero.
(This but for the 192/24 pair which seem to only differ at dither.)

I am not asking about what is audible. I am curious if there is any well-known workflow that makes this apparent lossy nonsense.

I don't have answers for why only a few findings re. 24/96 flac and wav. The straigt forward difference gives:

Code: [Select]

]$ sox -m -v1 "Derek Jones - Run With Me - 09 - Julie - 9624-FLAC.flac" -v -1 "Derek Jones - Run With Me - 09 - Julie - 9624-WAV.wav" -n stats 2>&1 | grep "dB"
Pk lev dB     -20.97    -23.96    -20.97
RMS lev dB    -52.19    -52.66    -51.76

Trimming one sample from wav file improves it:

Code: [Select]

]$ sox "Derek Jones - Run With Me - 09 - Julie - 9624-WAV.wav" a.wav trim 1s
]$ sox -m -v1 "Derek Jones - Run With Me - 09 - Julie - 9624-FLAC.flac" -v -1 "a.wav" -n stats 2>&1 | grep "dB"
Pk lev dB     -35.13    -37.83    -35.13
RMS lev dB    -66.37    -66.84    -65.95

Trimming 1.2 samples from wav file improves it even more:

Code: [Select]

]$ sox "Derek Jones - Run With Me - 09 - Julie - 9624-WAV.wav" a.wav rate $((5*96000)) trim 6s rate 96000
]$ sox -m -v1 "Derek Jones - Run With Me - 09 - Julie - 9624-FLAC.flac" -v -1 "a.wav" -n stats 2>&1 | grep "dB"
Pk lev dB     -50.31    -53.36    -50.31
RMS lev dB    -81.47    -81.93    -81.05

When put into DeltaWave, it can match it even better, to around -100 dBFS and the spectrum of delta looks like some kind of shaped dither:

Re: Any idea how these "same but different" masters were created?

Reply #2 – 2025-02-06 20:33:30

So a fraction of a sample. 1 sample offset was straightforward reducing the difference significantly, 1.2 samples improved over 1, I tried some wild resampling and got a good fit at 1.04 samples - since 48/44.1 is nearly 1.09, is it then plausible that one of these have been through a resampling to a multiple of 44.1?

(That means it is also not necessary to speculate whether one had been MQA'ed and un-MQA'ed, and the other not?)

Notice