Measuring objectively real differences in files with identical checksums

[wgscott]

Since it is now well-established by many members (see relevant PMs) that bitwise-identical files can sound hugely different, like night and day, what is the best way to get our measurements up to this new high standard defined by our golden-eared, golden-aged colleagues?  This is an absolutely imperatively important topic that really deserves to be interjected into every discussion at every opportunity.

[kumakuma]

I agree. I have received test files from the individual in question on several occasions and have tested them using every means available to me but I can't find a difference. I'd really like to know what I'm missing here.  

[wgscott]

If only Science had an answer.

[pkane2001]

12 minutes ago, wgscott said:

Since it is now well-established by many members (see relevant PMs) that bitwise-identical files can sound hugely different, like night and day, what is the best way to get our measurements up to this new high standard defined by our golden-eared, golden-aged colleagues?  This is an absolutely imperatively important topic that really deserves to be interjected into every discussion at every opportunity.

I understand that taking photos using a hand-held, low resolution, shaky camera off of a TV screen is the way to prove this scientifically. Despite my attempts to understand it, the proof still eludes me: the math is just too hard.

[kumakuma]

@sandyk

 

Chris has said that you are not allowed to post in this sub-forum. I have reported your post. 

[The Computer Audiophile]

If Sandy has objective data to provide, he is welcome. Otherwise, none of the subjective nonsense is welcome in the Objective-Fi area. 

[sandyk]

Chris

 It is highly unfair to permit members of this group to attack other members, without them having the right of reply

I should have the ability to at least correct blatant lies.

Alex

[The Computer Audiophile]

5 minutes ago, sandyk said:

Chris

 It is highly unfair to permit members of this group to attack other members, without them having the right of reply

I should have the ability to at least correct blatant lies.

Alex

Leave this thread. Nobody here wants to hear from you and your name isn't even mentioned. 

[Ralf11]

5 hours ago, wgscott said:

If only Science had an answer.

 

It does!

 

That area of science is called Abnormal Psychology.

[ASRMichael]

8 hours ago, wgscott said:

If only Science had an answer.

Yes like a black hole!!! 

[John Dyson]

12 hours ago, kumakuma said:

I agree. I have received test files from the individual in question on several occasions and have tested them using every means available to me but I can't find a difference. I'd really like to know what I'm missing here.  

i have narrowed down the differences to how my perception varies.   If a file is bit for bit the same, the the difference is in mood and perception.   My precision technical 'listening memory' is extremely short, on the order of 10seconds.   My 'hearing memory' for gross differences is a bit longer.

It is very easy to perceive the exact same material differently from listening attempt to listening attempt.

If I hadn't realized that memory problem in my own hearing several years ago, my project would still be dancing around in a random progression.  It is easy to get caught up in those kind of comparisions in informal discussion, but it is so very important to realize that hearing & perception does vary from minute to minute.

 

JOhn

 

[yamamoto2002]

Quote

Measuring objectively real differences in files with identical checksums

 

Identical checksum does not necessarily mean two files are the same bit by bit.

 

For example, MD5 checksum is 128bit fixed length integer value, therefore two larger than 128 bit files may share the same checksum value. Chances are low (slightly higher chance than of probability)

 

The following example files are created deliberately.

 

WAV files a.wav and b.wav in attached SameMD5Wav.zip share the same MD5 checksum value 161b403cd39f1f3f6cffb64b4dd8ccb9 but Waveform is different, beginning part difference is obviously seen.

 

C:\tmp>CertUtil -hashfile a.wav MD5
MD5 hash of a.wav:
161b403cd39f1f3f6cffb64b4dd8ccb9

C:\tmp>CertUtil -hashfile b.wav MD5
MD5 hash of b.wav:
161b403cd39f1f3f6cffb64b4dd8ccb9

 

 

 

SameMD5Wav.zip

 

[nugget]

1 hour ago, yamamoto2002 said:

 

Identical checksum does not necessarily mean two files are the same bit by bit.

 

 

There is a simple solution to this concern.  All you have to do is cross-check with multiple hashing algorithms to guard against a false positive from a collision due to weakness in one of the algorithms.  One can have high confidence in bit-for-bit equivalence if multiple hashing algorithms agree.

 

$ openssl md5 *.wav
MD5 (a.wav) = 161b403cd39f1f3f6cffb64b4dd8ccb9
MD5 (b.wav) = 161b403cd39f1f3f6cffb64b4dd8ccb9


$ openssl sha256 *.wav
SHA256(a.wav)= b4c9cba57fbe8150943b0d6c361af76bd48ba128f606be295f3a8cbbbba93af4
SHA256(b.wav)= cb6b1a94718fbb53b28ad99f9fc35a9c9d9f7534485c1605ac329bb04e471399

 

$ openssl md4 *.wav
MD4(a.wav)= 8de0b4d206a36321a439c9c150231a73
MD4(b.wav)= d78d9a340f322ba5749565db1fb0068f

Here we see that both MD4 and SHA256 reveal that the files are not equal despite the hashing collision with MD5.

[yamamoto2002]

30 minutes ago, nugget said:

 

There is a simple solution to this concern.  All you have to do is cross-check with multiple hashing algorithms to guard against a false positive from a collision due to weakness in one of the algorithms.  One can have high confidence in bit-for-bit equivalence if multiple hashing algorithms agree.

 

$ openssl md5 *.wav
MD5 (a.wav) = 161b403cd39f1f3f6cffb64b4dd8ccb9
MD5 (b.wav) = 161b403cd39f1f3f6cffb64b4dd8ccb9


$ openssl sha256 *.wav
SHA256(a.wav)= b4c9cba57fbe8150943b0d6c361af76bd48ba128f606be295f3a8cbbbba93af4
SHA256(b.wav)= cb6b1a94718fbb53b28ad99f9fc35a9c9d9f7534485c1605ac329bb04e471399

 

$ openssl md4 *.wav
MD4(a.wav)= 8de0b4d206a36321a439c9c150231a73
MD4(b.wav)= d78d9a340f322ba5749565db1fb0068f

Here we see that both MD4 and SHA256 reveal that the files are not equal despite the hashing collision with MD5.

 

This sure reduce the chance of hash collision, two larger than 512 bit files may happen to share the same MD5 and SHA256 and MD4 checksum value approx in 

of probability.

[jabbr]

I tried to listen for differences in sound from two but identical files, but couldn’t hear any. Then I realized that my file system (ZFS) automagically maps identical files to the same location (actually it does block level de duplication) , so I can’t even test this. 

[Archimago]

5 hours ago, jabbr said:

I tried to listen for differences in sound from two but identical files, but couldn’t hear any. Then I realized that my file system (ZFS) automagically maps identical files to the same location (actually it does block level de duplication) , so I can’t even test this. 

 

Well then, that "version" stored on your file system could be the absolute best or absolute worst copy of the music...

 

Which is it based on the theory of how this is supposed to work? Or is it all just "different" and equal with no relative judgment possible!?

[yamamoto2002]

About 20 years ago of issue of local computer magazine 月刊アスキー, there was an article about sound difference of identical DAW project file stored on different hard drive.

One drive is IDE HDD and another drive is SCSI HDD, and a music composer says SCSI HDD sounds better.
On the next issue, the cause of sound difference is found, it is caused by sound stutter due to slow drive transfer speed of IDE HDD

[pkane2001]

13 hours ago, yamamoto2002 said:

 

This sure reduce the chance of hash collision, two larger than 512 bit files may happen to share the same MD5 and SHA256 and MD4 checksum value approx in 

of probability.

 

The funny thing about hashes is that for bit-perfect analysis, you don't need to use them

 

Just compare the files, byte for byte, or sample for sample. Easy, and no confusion. Ever.

[jabbr]

6 hours ago, Archimago said:

 

Well then, that "version" stored on your file system could be the absolute best or absolute worst copy of the music...

 

Which is it based on the theory of how this is supposed to work? Or is it all just "different" and equal with no relative judgment possible!?

In all seriousness files stored on media could certainly, no, will have different levels of readout noise. That could be audible. However, the readout noise is not preserved with file copy. In the case of ZFS, the data is also mirrored, and read into RAM cache. In my case, from RAM cache, it is transmitted across network so whatever is going on the NAS is entirely electrically isolated from the streaming endpoint. My network is 10G fiber and assuming my switch meets the compliance testing standard, there is extremely little noise in the bits. Of course there is zero common mode noise  

 

Absolute best or worst stored in my magnetic media, as long as the bits are readable, the bits sent to my DAC do not reflect this noise with virtual certainty (we have no reason to invoke entanglement 😂) and this issue is moot for me.

[jabbr]

On 2/17/2020 at 8:41 PM, kumakuma said:

I agree. I have received test files from the individual in question on several occasions and have tested them using every means available to me but I can't find a difference. I'd really like to know what I'm missing here.  

It’s really quite easy if you have the equipment. Make two copies on a CD and then magnify and look at the pits. They will be different. Or measure the magnetization, again different. 
 

At a fine level all digital bits are represented by analog processes and the analog is different from bit to bit.

 

(of course digital systems are explicitly designed to deal with this)

[pkane2001]

2 minutes ago, jabbr said:

It’s really quite easy if you have the equipment. Make two copies on a CD and then magnify and look at the pits. They will be different. Or measure the magnetization, again different. 
 

At a fine level all digital bits are represented by analog processes and the analog is different from bit to bit.

 

(of course digital systems are explicitly designed to deal with this)

 

I prefer to use a tunneling microscope. Makes the differences much easier to see 

 

[plissken]

18 minutes ago, pkane2001 said:

 

I prefer to use a tunneling microscope. Makes the differences much easier to see 

 

 

I'm sure the 7nm lithography they use to make IC's matters too. Best get one cherry picked from the center of the wafer.

[jabbr]

Just now, plissken said:

 

I'm sure the 7nm lithography they use to make IC's matters to. Best get one cherry picked from the center of the wafer.

Snark aside: 

 

1) are there differences in bit-identical files?: yes, always

2) are the differences meaningful for SQ in my system?: no, never 

[plissken]

4 minutes ago, jabbr said:

Snark aside: 

 

1) are there differences in bit-identical files?: yes, always

2) are the differences meaningful for SQ in my system?: no, never 

 

It's not meant as snark. Manufacturers will bin and specialty sku parts from the center of the wafer vs the edge.

 

Will that mean differences for audio? No. Never. In no ones system.

 

It would be absurd knowing what we know to suggest otherwise. Both of us know better. 

[kirkmc]

On 2/18/2020 at 1:35 AM, wgscott said:

Since it is now well-established by many members (see relevant PMs) that bitwise-identical files can sound hugely different, like night and day

 

Is it?

 

On 2/18/2020 at 2:23 PM, John Dyson said:

i have narrowed down the differences to how my perception varies.   If a file is bit for bit the same, the the difference is in mood and perception.   My precision technical 'listening memory' is extremely short, on the order of 10seconds.   My 'hearing memory' for gross differences is a bit longer.

It is very easy to perceive the exact same material differently from listening attempt to listening attempt.

 

Of course. The way you hear anything varies according to so many different conditions and filters that it's pretty hard to think that it is ever exactly the same. The time of day, your fatigue, the conditions of your sinuses, whether or not your blood sugar is normal or low, and the level of alcohol in your blood; all these things affect what you hear. 

 

I find these discussions fascinating, because of the attempt to explain by subjectivity something that doesn't exist, yet it's something that people can hear. In other words, music can sound different at different times, but assuming that two different bit-identical played through the same equipment sound "hugely different" for any reason other than that of the listener shows a real desire to complicate things. 

 

The second quote above should be so obvious, yet some people cling to the far more complicated postulate. Occam's razor, and all that.