Thursday, 30 December 2010

Digital audio myths give me jitters

Now, i'm going to say right out - the subject i'm about to rant on is highly controversial. The field of digital audio is much too polluted with ghost stories, common misconceptions and outright bullshit. I am also going to say right out that by no means i am an expert in that field, and by no means i assume that i can't be wrong. However, i am in such a position of having knowledge and view of things from a number of perspectives.

The following article is a collection of musings based on this article by Roger Nichols and this piece by John Vestman.

With all due respect guys, for the most part this is complete bullshit.

First, let's take a look at Vestman's piece. Leaving jitter issues aside for a moment, this guy seems to genuinely believe that 1-to-1 perfect copy of the file is somehow not identical to the source? Moreover, i would say that in the digital domain there is no such thing as "original file" and "a copy" - they are identical! They are both original! They can't be different because it is the same file, they are effectively the same data, down to the last fucking bit!

I can understand why this shit even comes to their heads. These guys are audio engineers. They are used to analog, they are used to physical, they are used to how things work in real world. They are used to the fact that two pieces of the same hardware can be different, they are used to the fact that two copies of vinyl are not necessarily identical, hell, they are even used to the fact that digital copy from a digital tape might not be the same as the source.

However, guess what, these things, while they might be operating in the digital domain, in fact are analog. The digital tape is analog media, modified to hold digital data. CD's are not analog media. Hard drives are not analog media. They are digital media. And while it is possible for digital copy of analog tape to be different from the original, it is not possible for digital copy of digital media being different from the original. The copy is either identical, or it is not a copy. This is what audio engineers usually don't know, and this is exactly what we, programmers, do know. We programmed the thing, goddammit, we know it is perfect, because in binary world it is either perfect or it does not work.

I might agree that playing the same data from different source (IDE harddrive, FW harddrive, CD) could sound different, but this has nothing to do with the media. It might have something to do with the noise of CD player, or different head position during listening (comb filtering effect), or maybe speakers picking up different noises from wires, but it is certainly not the media itself. No CDR's can have "solid and balanced high-to-mid-to-bottom and wide sparkle", no hard drives are better than the other. I'm sorry, Mr. Vestman, i respect your work and your experience, but this is complete bullshit.

Now, moving to the jitter, error correction et al. Again, audio engineers tend to think analog. They tend to think that these microscopic bumps on the CD are actually representing sound. They don't. They are just numbers. It is true that the recorder jitter might put these bumps in not-so-perfect distance from one another - that's where the pure digital world meets analog reality.

However, that does not mean that once these bumps get read back into the memory, they retain these imperfect distances. There is no such thing as "jitter" inside a computer memory, and this is exactly what happens when the CD gets read. Unlike vinyl, these bumps don't make up the sound themselves, they merely represent 1 or 0. Hence, when the data is read, it is read not giving a flying fuck about jitter, because it is not fed directly to the sound device, it is read into a buffer! And when the data reaches the DAC, it has already been aligned properly - there is no jitter at that point! Same goes for the part II of that article, where Mr. Vestman goes on about moving parts of HDD introducing jitter and sound data being lost due to calculations on the digital data... The jitter part is obvious bullshit, and the part about calculations is... erm... was true, back in the day of 16-bit DAWs, but now that DAWs are 32-48-64 bit resolution, these errors are still there, but they are undetectable even by measuring, let alone by ear. It's hilarious to read a highly technical article from a guy who has no clue about how things work.

There is another thing that gets mentioned - imperfect media and imperfect playback devices. Imperfect media does not introduce jitter, that is complete bullshit and i have already showed that. What it can introduce, however, are errors. Now, read errors are not as common as John Vestman tends to think - quite on the contrary, they only start happening after your CD (DVD, HDD, thumb drive) have been to hell and back. So obviously, these errors happen rarely (with good media of course, if you're a cheapskate - you get what you pay for) and are not that much of a big deal.

There is an interesting phrase by Mr. Vestman - "Error correction helps, but when it comes to your master, you want the ultimate... not second-best". Obviously, Mr. Vestman again thinks analog. In the analog world, when you restore a signal that has been damaged, you can never get the original signal back - that's just how it is.

However, in the perfect digital world, it is fucking possible. It is possible to damage your data, then use error correction and get your data back in perfect condition. Obviously, Mr. Vestman has never used WinRAR and its recovery records to recover a broken archive. Yes, i understand that this is slightly different, but if what Mr. Vestman is talking about happened in real life, it would be a disaster because it would not be possible to copy files on CD without fear of them becoming corrupt. If those "error correction codes" were only approximations and not the perfectly recovered original signal, that would work on audio but would totally ruin data CD's. And since data CD's and music CD's are essentially the same media - one must be a fool to think that error correction works flawlessly on data CD's and doesn't work on audio CD's.

In other words, these errors are perfectly recoverable. It is not like noise reduction, the correction process actually restores the original data. Now, it is true that some errors are unrecoverable and lead to guessing and/or read errors, but these usually involve media being damaged or worn out, which is not the fault of the digital system anyway.

A couple of words on jitter. When does it happen? When it is physically possible for jitter to happen? It can happen only at one stage - at the boundaries between analog and digital. When you digitize your signal, you might introduce jitter. When you turn your signal to analog, you might introduce jitter. That's fucking it. There is no jitter in reading the digital data into a digital container. There is no jitter in processing digital data. There is no jitter anywhere other than in the process of converting digital to analog and back.

What can introduce jitter when playing back the CD? Poor DACs in your CD player. That's fucking it. There is no way jitter could be ever introduced in other places or on other steps of the process of playing back the CD, simply because these processes don't involve anything that introduce even the remote possibility of jitter.

So, i am sorry to say that, but even great minds can make mistakes. Mr. Vestman, you're wrong and you're full of shit. Same goes for Roger Nichols. I'm with Ethan W(h)iner on this one.

Some people say that audio engineering is art, and not everything can be scientifically explained and/or verified by tests. I say it's complete bullshit. Audio engineering is as much engineering as it is audio, and you guys should understand that and know your math.

Differences? Great way of determining the differences is the null/summing test. Not practical for audio equipment? Double blind test (although this guy tries to make a point that blind tests are flawed). That's it. Pure science. If you can't reliably hear it - it's not there. If you can't see it on an spectrum analyzer - it's not there. If things null - they are the same.

Regarding the article about blind tests being flawed, i would argue that the guy might not even heard the artifact, he might have discovered that by running the signal through spectrum analyzer - that 1.5Khz tone could be easily spotted. I also might argue that the test conditions were not adequate to perform a good blind test. It should be noted that it has to be done right too. All in all, the guy makes some valid points, but comes to the wrong conclusion. What he should have concluded is not that "double-blind tests suck", but "double-blind tests should be done right". In other words, as any scientific test, the double blind test should be done at the most ideal conditions possible. Basically, the comments to the article say it all.

A lot of people claim they hear a difference between two amplifiers, between two pairs of headphones, between two cables, between two X, between two Y... I'm not saying they can't be different, i'm saying that sometimes it is just not physically possible for the difference to be there, or to be noticed. I agree headphones sound different, but i do not agree that, say, DAWs sound different. Because math is math. If 1+1 is somehow not equal to 2, then your host is doing it wrong. If 1+1 is 2 but you hear something else - then it's something wrong with either your ears, or your listening equipment/environment, or this, or that - there is a gazillion of factors that could influenced your perception. Our ears aren't perfect, and they can be easily fooled - so we should rely on science to prove things instead of just subjective experience.

No comments:

Post a Comment