But in fact we're not talking about that. We're talking about 2 identical digital files behaving differently at the same time. The controversial assertion is that the way the file is generated, not the data in the file itself or the way it is played, causes it to sound different. It's ascribing some "magical" quality to the data in the file that somehow remembers how the file was made independent of the data in the file.
The Absolute Sound (magazine) take on many aspects of computer assisted music reprodu
- Thread starter rbbert
- Start date
But in fact we're not talking about that. We're talking about 2 identical digital files behaving differently at the same time. The controversial assertion is that the way the file is generated, not the data in the file itself or the way it is played, causes it to sound different. It's ascribing some "magical" quality to the data in the file that somehow remembers how the file was made independent of the data in the file.
Sorry, it is not the "same time". Unless you read it all to some memory buffer and the computer is idling meanwhile the data is being accessed from the disk in real time. As you say, it is unlikely that any thing can be changed because of this difference, but I should point again exactly the same thing happens with optical drives - they have memory buffers and the controller reads from the memory, not from the laser output. .
Of course the way the files were generated has nothing to do with it in a conventional sense. But, and a big but, a difference in the temporal history of the generation of the file may cause it to be laid down on the hard disk in a significantly different fashion: most plausibly, how fragmented the file is. If one version is written so that it is far more split up into chunks all over the hard disk, then the hard drive is working harder to read the data on playback, the electrical activity in the power supplies is at a much higher level, and finally, way down the chain, more subtle electrical interference is injected into the audio, analogue circuitry ...
Frank
I'm sorry to disallusion you, but operating systems are obsessed about splitting up files on hard drives!! You have to work very, very hard to stop files, especially large ones, many megabytes in size (audio, anyone?) from being shredded into multiple little bits all over the place. As soon as you add and delete, add and delete, add and delete some decently sized files of any type your hard drive starts to look like a tapestry. Hence Micro's comment about defragmenting, which may or may not work ...
Frank
Frank
OK, perhaps that explains why I hear no difference, because I've been copying the files to a freshly formatted USB flash drive and playing back from that ...
But that certainly has no support for the explanation that ripping speed is the only difference between the files
But that certainly has no support for the explanation that ripping speed is the only difference between the files
Yes, that is what the people doing the test propose, but I would suggest that if you actually examined the signal on the hard drive at the analogue, not digital, level you would see subtle differences. To give an extreme example, there is almost no such thing as truly deleted data on a hard drive: even overwriting it many times doesn't do the job. If somebody really wants to know what was stored on your disk historically they can set it up in a very sensitive measuring environment and pick up the remanents of what was stored many generations ago. Ultimately, everything can be looked at as analogue. And, from another angle, everything can be viewed as digital. Sort of like light: is it a particle, or is it a wave?
Frank
The subject of read head electronics aside (whilst very interesting), I thought the issue was if the same file, stored in memory and provided to the DAC, could sound different? If the data is bit-for-bit there should be no difference. Are we now arguing the data delivered to the DAC is not identical due to the reading mechanism? Just trying to see where all this stands... Obviously there are potentially differences in sound if the data is not the same at the DAC. I am not sure the base file is to blame in that case, however. - Don
Don, I wouldn't be arguing that the data fed to the DAC is different if considered as a digital signal. But if you looked at it as an analogue input to the DAC circuitry it may have significant differences: noisier, more jittery. Where the reading mechanism comes into it, is that if it's working harder then the circuitry which ultimately transfers the musical data to the DAC has a greater chance of causing unwanted electrical interference to be added to the signal, as an analogue "component", because the CPU and associated circuitry is more active.
Frank
Frank
Lost a long post, I hate this computer...
I understand read channel electronics and (to some extent) data correction and recovery. What I do not understand is how two identical files could sound different. Random artifacts will change with each playback; I suppose a string of trials with the two files randomly interchanged might average that out. Deterministic (data-dependent) artifacts should be the same for the two files.
This has the hallmark of another religious thread I'd best avoid... - Don
I understand read channel electronics and (to some extent) data correction and recovery. What I do not understand is how two identical files could sound different. Random artifacts will change with each playback; I suppose a string of trials with the two files randomly interchanged might average that out. Deterministic (data-dependent) artifacts should be the same for the two files.
This has the hallmark of another religious thread I'd best avoid... - Don
I think this has been mostly addressed but I will walk us through it.
The process of creating files involves segmenting it and storing it one block at a type. Block sizes can vary but for now, a common size is 4096 bytes or half as many PCM samples (since each is 16 bits).
In the process of ripping the CD, the OS is asked to find a free block on disc for each 4K chunk above. At the absolute, the OS is free to pick any random free block it wants. Doing so reduces performance so there is usually an optimization step where an attempt is made to find a free block that is closest to the one previously used as to make the file as contiguous as possible. But no assurance is provided. When people try to benchmark hard disks and such, they always start with a freshly formatted drive as to make the file allocation more predictable. But even that, is not a safe bet as background activity can cause files to be written that create fragmentation.
So it is safe to say that there are differences in block allocations of two identical rips on the hard disk on a normal system as used by the authors.
At some time later, we try to play those files. The Media Player will allocate memory to read the content and will ask the operating system to read that chunk of the file for it. The OS as I believe was mentioned, does that but also reads ahead, anticipating future requests from the same file. This is done because the drive keep spinning and if you don't read the block sequentially, you wind up "missing a revolution" causing you to wait a while before the block becomes available to be read by the stationary drive head.
How much the OS pre-reads is unpredictable. It will examine for example how much free memory it has.
Once the data is read into the OS buffer, it will then be copied to the application buffer, which will in turn pass it on to the audio driver (S/PDIF in this case).
In a multiple-trial blind tests, the second read of the file will most likely find all the blocks already in the OS buffers and no disk activity occurs. But the copying of the file buffers will go on. The disk drive will continue to spin but not perform any other operation *if* there is nothing else going on in the system. Alas, there is always a ton going on in the system. You will be horrified at how many disk read and writes are going in your system even though you are not doing a thing! There are apps you can download that show you this.
What the above means is that the behavior of the system in this manner is rather chaotic and unpredictable.
The notion that a digital file sounds different from another relies on two aspects:
1. The timing of digital samples has changed. The authors used S/PDIF interface on the PC motherboard. Read another way, they lowest quality digital interface they could find. So we can assume, if timing is going to change, it will.
How will timing change? It will change because the clock circuit powering the S/PDIF interface is under constant attack from all the PC activity. Every drive, CPU, or GPU activity is likely to create pulses on the power line in addition to leaking RF and such onto the S/PDIF clock. Such variations could range from making no difference whatsoever, to some difference. Without measurements, we are in the dark as to where we are in this.
Timing could also change if the S/PDIF waveform changes under load.
2. Electrical coupling. If you connect two devices using S/PDIF, you now have a common ground between the PC and your audio gear. This means that noise and such can bleed into the DAC and cause loss of accuracy, ground loops, etc.
So the theory of why something could make digital audio sound different when fed from identical PCM samples is sound. What is not sound is that by doing X, you reduce its impact. The OS as explained, is a random and unpredictable animal. It is entirely possible for example that when I don't read from the disk drive, I create more jitter or noise because the CPU is processing the data faster than if it idled between disk reads. On the other hand, maybe those X millisecond delays to read a block from disk creates a jitter at that frequency and that jitter is more audible than the random one created by the CPU.
Importantly, as I mentioned, if we are down to the level where identical files located in different spots on the hard disks making a noticeable difference, then all bets are off. You can count on the songs in one album to have varying levels of fidelity even if you ripped them with the app you thought was the best!
And let's dispense with a clear untruth: there is no way you should prefer one ripping application vs another when they both create identical bits! The ripping app's job is to create the bits, not to play it. If both apps create the same bits, then they are identical. If the outputs sound different because they are located in different parts of the hard disk, that has nothing to do with which one is a better ripper. It is simply by chance that you got the bits in different place. And in song after song you are going to experience the same anyway.
This whole issue is such a bizarre rabbit hole that I just like to avoid it altogether by using digital audio interface that deals with both #1 and #2 above. Once there, there is no worry about these factors and you have a reliable and performant system. The opposite of this approach is to use on-board S/PDIF interface and then have to spend good money to try to still get good performance out of it. Why not start with a much cleaner signal that is immune to PC randomness?
But, if the two identical files are stored differently, that is, the electronics responsible in every sense for reading and passing on the nominally digital data to the DAC have a different "journey", their activity varies through the whole exercise. We are not talking artifacts here, we are talking the language of short wave radio reception here: every, every tiny thing will affect how clear the reception is in that case.
IF, and a mighty big IF, the DAC is "perfect", and then of course no matter what sort of interference you throw at it, from noise on the digital input, RF interference, noise on the AC mains in to the DAC's power supply the 2 files will sound the same. Of course, if you're an objectivist you know all audio components are perfect, their spec. sheets say that, so it must be true ...
Frank
IF, and a mighty big IF, the DAC is "perfect", and then of course no matter what sort of interference you throw at it, from noise on the digital input, RF interference, noise on the AC mains in to the DAC's power supply the 2 files will sound the same. Of course, if you're an objectivist you know all audio components are perfect, their spec. sheets say that, so it must be true ...
Frank
Thanks, Amir, for that excellent writeup: nicely details the "mess" you get into if you really want to understand what computers "do", whether as a PC, or music server ...
Frank
Frank
No, Don, I think you're right. Even if there are consistent read differences from the hard drive for the two files (which in itself is not very likely), the better music players read from memory, not from the hard drive. In fact that's often proposed as the biggest reason they sound better than something like iTunes. They also try to turn off as much computer background activity as possible while playing, to decrease the CPU and memory activities.
Last edited:
Thank you, Amir!
And yet, they claimed that using their (Samsung) optical drive with this setup, the sound was better than the PS Audio PWT using I2S into the same DAC (PS Audio PWD)??
And yet, they claimed that using their (Samsung) optical drive with this setup, the sound was better than the PS Audio PWT using I2S into the same DAC (PS Audio PWD)??
Yes, I am dubious that the on-board S/PDIF had such a performance in two systems no less.
Jeff Fritz has just posted in another thread a link to an interview with Matan Arazi on Music Servers
http://ultraaudio.com/index.php?option=com_content&view=article&id=135:matan-arazi-on-music-servers&catid=31:general&Itemid=46.
It refers with detail to many of the aspects that we are talking about in this thread, including the bit perfect concept.
http://ultraaudio.com/index.php?option=com_content&view=article&id=135:matan-arazi-on-music-servers&catid=31:general&Itemid=46.
It refers with detail to many of the aspects that we are talking about in this thread, including the bit perfect concept.
A very good read, micro, thanks for that. The bloke's on the money, has all the right ideas: as he points out, the bit perfect aspect is a very minor starting point, the guts of it is an obsession about minimising the creation and transference of interference to the analogue audio side of the circuitry, and why laptops, etc, are a bad idea ...
Frank
Frank
Excellent indeed, and the part about dedicated music servers as distinct from a PC or Mac mirrors my own thinking on the matter. I definitely get better sound from a USB flash drive into my Oppo than from any of my computers, no matter how I hook them up to my DAC. People rave about the Bryston BDP-1 for the same reason, I suspect.
His systems knowledge is quite good. Unfortunately, I don't think he understands Psychoacoustics and its application to jitter. He also makes a scientific mistake: assuming something that makes sense to a layman but he has not bothered to prove: "less is more."
Less is more resonates so well with audiophiles. We all get the idea that if we ran audio through 20 analog copies, it would degrade. So we automatically assume what he is saying is true: less traffic in your computer, the better.
Alas, the above is dead wrong when it comes to jitter. Jitter causes distortion sidebands. If you are playing a 3 Khz tone, and have jitter that is at 1 Khz, you get distortion tones at 2 KHz and 4 Khz. Looking at the Fletcher-Munson ear sensitivity curves, we see that the distortion products land where our ears are most sensitive:
Now let's double the system activity and jitter to 2 Khz. Now I am getting distortion that is at 1 Khz and 5 Khz. We see that those two sidebands fall further out in the curve where the ear is less sensitive relative to 2, and 4 Khz. Let's triple the jitter to 3 KHz, and now we have the left sideband disappearing and the right one at 6 Khz. Clearly now we have even less audible jitter.
Here where it gets fun. Let's say that we have 200 things going on in the PC, and all doing lots of different things. Now instead of having those clean jitter frequencies, we get random hits on our clock. Random signal has infinite components and hence creates sidebands at ALL frequencies. Since jitter now exists at all frequencies, it once again becomes noise! Since noise is just noise, it is not a contributor to system distortion. So 1000 times more activity is preferable to 10 activities that have clear rhythm.
Of course, one could come up with sample data that works the opposite of what I have shown. Take jitter at 100 Hz vs 100 Khz. 100 Hz jitter creates sidebands that are very close to our original frequency (i.e. +- 100 Hz). This gets into another bit of psychoacoustics called masking. This says that if a signal is lower in amplitude but very close to another signal, we won't hear it. Such is the case with 100 Hz jitter. BTW, this is the reason flutter in tape and LP are not audible even though their amplitude if far above jitter in digital systems! The distortions created are masked so heavily that they are much less audible.
Here is the graph showing masking:
The sideband in red is clearly more audible than the one in blue.
This is why I said this is such a chaotic problem. Less is more sometimes. Other times, more is really more!
Similar threads
 | Steve Williams Site Founder | Site Owner | Administrator |  | Ron Resnick Site Owner | Administrator |  | Julian (The Fixer) Website Build | Marketing Managersing |