DSP = Digital Signal Processing
I was going to make another point about what wes480 was commenting about the "destructive" DSP per se. How the sound quality DEGRADES as DSP is applied.
Yes, it does degrade, but not quite in predictable ways when compared to analog devices. I will not go into specifics in the comparison, I will just share a few thoughts about DSP and bit depth.
When you sample audio on a A/D converter, the incoming voltage to the A/D COULD be varying very quickly, and by quite a bit. It usually is with "music". The sounds instruments make are made of of several frequencies combined, and that is what the "timbre" is. "timbre" give us the distinct tone of the instrument. So, a musical "note" is actually a very complex wave form. Not anything like a square wave which would look something like:
__
_l l__ "really bad eh"?
you see a "flat top" to the amplitude of the frequency. A "pure tone" (meaning a tone that does not contain any other frequencies with it) would look something like that, or a very distorted tone.
But a complex musical "note" would look more like:
www (sort of)
The volume would go more up and down because the combined frequencies are doing all sorts of wonderful things to the audio, and at varying amplitudes!!!
So, in a complex musical note, certain frequencies would die off faster than others, others would sustain longer than others, etc....
So, this is why audio goes up and down in volume quite a bit! It really does, unless you DISTORT the input to something, which would sort of cut off the peaks of the audio. This cutting off of the peaks means that our ear cannot really get a good handle on the HIGHER FREQUENCIES because the never fully develope to a peak. So the high frequencies tend to sound more vague, sometimes described a "smoothed out".
Be patient here guys, this will make a lot of sense in a while...
Now, our audio is going up and down, up and down, up and down in voltage SMOOTHLY, CONTINUOUSLY. It looks sort of like:
~~~~~~~~~ (up and down being volume, left to right being frequency...get it?)
BUT, the A/D converter can only store so many values!!! Oh no!!!
What happens if when the converter "samples" (the sampling rate....) that the current incoming voltage falls between two certain points that the converter can store a value? Say the converter can store 6 and 8. But the incoming voltage is 7! Yikes. What to do!!! Digital doesn't like to think!!! It wants to be told what to do!!!
The voltage is rounded up or down (from what I understand, usually up) to fit the next available storage value. This shift in voltage causes distortion to the audio, and is referred to as quantinization errors. It is mostly appearent in low volume audio, and to tell you the truth, I don't remember why, or understand log scales and the A/D circuit design well enough to explain why, but I guess it has to do with there being less bits available to store the incoming voltage values. Bummer....
So, the lower the incoming volume gets, the less bits that are available to represent it!!! So, the quantinization errors are WORSE on incoming low voltage audio than high voltage audio (high voltage being a relative term ONLY to voltages that are lower than it...we are talking milliamps here guys!
)
Okay, so the incoming audio is distorted somewhat when it is stored. So we want to keep as much of our audio above the point where we really start getting bad quantinization errors.
The theory behind 24 bit audio is that those quantinization errors are moved MUCH lower than what out hears can hear. Now this really doesn't mean much to us yet. 16 bit audio, when done well, with a good circuit design, has proved to work pretty good for most POP music. Orchestral recordings tend to sound "good enough" if the noise floor of the preamps, mics, etc. are low enough and not contributing their own nasty distortion. But sometimes, the RIGHT kind of continuous noise can help hide these quantinization errors because the noise is a constant going to the converters, so a constant voltage is applied, the blending of this noise and the music content BELOW it keep voltages stable, so bits aren't being toggled on and off, etc....blah blah blah...Dither is just noise applied and does this. Some recordings don't REQUIRE dither because they are so noisy that you can't hear the quantinization errors anyway.
I regress.
So, here we have our audio stored in fixed voltages in the digital domain. It plays back cleanly enough. Certainly no worse than it was stored IF we don't do anything to it. The fixed values that can be stored are stored. The audio is "perfectly" stored, and will not degrade because it is played back (unlike analog tape which "degrades" EVERY time you play it...you just don't hear it until the next time you play it...but actually it takes SEVERAL passes of tape before you really hear a bad difference...I regress!!!damnit!).
So we haven't done anything to the audio in digital and it sounds like it should. Cool!
When you apply a fader move though, problems start!!! The voltages have to be "recalculated" because of the fader move. If you turn the audio down .1 dB, ALL of the audio must now be turned down .1dB. Yikes!!! Now all of our fixed stored voltages are messed up damnit!!! So, all those voltages that fall between two storable values must now be rounded up again! MORE DISTORTION FROM QUANTINIZATION ERRORS!!! YIKES!!! So much for our "perfectly" stored audio now eh?
You can see how many "steps" of DSP can start really making a mess of things. You turn down the volume on the fader a bit, Quan (quantinization...that is too long to write, so I will say Quan from here on out...whew! thanks for understanding....
) errors there! You apply an EQ to the sound, yup, more Quan errors. You want to compress it? Ok, but yet even MORE Quan errors!!! On and on it goes. Where it stops everybody knows! (heh...I am funny right now...) It stops ONLY when you stop warping the audio with DSP! Oh yeah, that audio that you fader moved, eq'ed, and compressed, it is contributing to the Master Buss, which will have to calculate ALL of the tracks combined into some kind of voltage. Oops, another Quan error there! Oh, you want a compressor over the whole stereo buss now? Dang, yet ANOTHER Quan error. On and on...it keeps on crying....
Makes me wanna cry how many times DISTORTION was introduced to the audio from the time I started mixing with DSP, until I burn a master disk! Digital is perfect? LOL....yeah right!!! Far fucking from it!!! All that distortion introduced add's up to a sound that is quite a bit harsher and with far less depth than the original sound. The stereo spread seems to shrink up (more when the DSP is really having to honk!). The highs start getting crackly. Yucky audio!!! Thank you DSP!
Okay, now, it is NOT as bad as I made it out here really. A lot of that hash is in the lower bits. Remember, Quan errors seem to effect low level audio far worse than high level audio.
24 bit theoretically gives us the ability to move the Quan errors down to a level where our ears cannot hear it. Yeah? you say? Cool! If the software is written well, floating point DSP, or very high internal bit depths can used so that we can save Quan errors until later. Coolness, maybe. Did the software authors WRITE the code to work well? All those floating point calculations take a BIG toll on your processing power, and to keep within the Windows 32 Bit Floating Point DSP scheme, they would have to spend more CPU cycles to increase this internal bit depth of the software, or just say "screw it, these morons won't hear the difference! They are using Delta 1010 cards!!!!" (okay, they probably say SoundBlaster Lives..just thougt I would take a poke at you Delta Losers...errrrrr, I mean Users....
). So, I believe that in a lot of cases, rather than writing out GOOD code, that will deal with all these issues of recalculating the audio adequately within a 32 bit float environment, and STILL not tax your processor too much with EXTRA calculations to have BETTER code and/or deeper internal bit depth, they probably just go ahead and allow "truncating" to happen, or in possibly a better scenario, "dithering" to be applied. I suspect truncating because dithering is yet another hit on the CPU, which is already taxed quite hard by all this other processing!
Are you seeing the story now? Your Wintel PC has some lame ass slowmo processor that has to do all the Windows task, AND run the software!!! Sure, you could just say "Hey, just apply great processing to the audio at any expense, and I will happily wait for it to process if it is going to be better!". But what about REAL TIME MONITORING of the DSP you apply? If you start taxing your CPU too hard, your audio WON'T play. Sooooo, I suspect that software writers try to write stuff that isn't too bad on the CPU. Ooooopppppsssss!!! A compromise was made! Crap, the DSP is as good as it could be by a long shot! Damnit!!!
So, there is the problem with DSP. Unless you have dedicated processing to allow for more powerful code to be written, well, more powerful code WON'T be written. That is that, and that is what you have to deal with on a Wintel recording solution.
The reason why something like ProTools usually sounds better is that the processors in their box that does the DSP (yup, a real protools system DOESN'T use the computers processor do to anything but send commands to the DSP box that comes with ProTools that DOES the DSP) is DEDICATED to the task! No OS functions to perform. No hard drive saying "hey, I need to report my thermal calibration here so I can keep going...excuse me...". Nope, you have a powerful processor JUST for processing the audio. Cooooooooooooool! When designing a system like this, you can say ONLY a certain amount of processing can be ran per processor! Coooooooooooool. The processor can handle more powerful code without the worry that you will expect it to run more than it can handle. That is what those DSP Farm Cards are all about. THEY supply the dedicated processing power!!! Each has it's own processor. It is a dedicated processor capable of powerful code! Coolio!
So, if HARDWARE could handle MORE POWERFUL CODE, then more powerful code could be written!
But, hardware comes at a price! AND...software takes a long time to write. So, STANDARDS have to be kept for something like this to work. Of course a software company can write code that requires the processor to use two cycle to process what another software would do with one cycle tick. But now we get into performance issues! The software company doesn't want you to NOT use their processor because it taxes your system right? Hell no. That is suicide. People buy it to use it! So, instead of lossing out that way, they make the code less powerful! Ever noticed how a Waves plug in taxes your CPU sometimes 3 or 4 times as much as a native plugin to an mixing application? Notice how it sounds better? But, you can't expect your little ol' 900MHz processor to handle 32 of those high powered plugin's at once can you? Along WITH editing you did on the track, automation, effects, etc etc etc.....So, Waves can only go so far in their processing. Their latest version is as far as they go until the totally rewrite the software to use more cycle ticks. They have to wait until processing speeds for the AVERAGE user are high enough for them to go ahead and use more powerful code to do their thing. So indeed, DSP get's better all the time (new versions of it at least...).
But, I am STILL very leary that you can get adequate DSP that starts rivalling even decent analog processing.
The nice thing about analog is you usually are only dealing with the gears self noise and the buss summing issues. In digital, well, no self noise, but you still have buss simming issues (combining volumes together...) as well as Quan errors for EVERY single thing you do with DSP. To my ears, these Quan errors, along with inadequate code being written, equals audio that is just not all that great. It sounds like it has a veil over it. An upfrontness to the sound is missing. The more DSP applied, or the higher the track count, the worse it it! In analog, self noise is about only bad issue about high track counts, and good circuit designs can go a LONG ways towards eliminating self noise (that is why killer sounding analog console can set you back half a million dollars! they are freakin' expensive to build because the circuit designs call for components that are expensive!!! but this circuit design is LOW NOISE, and has the benefits of a very large operating range, and doesn't impart bad sounding artifacts to the sound passing through it....etc etc etc....).
So, there is the story as I see it. I am very open to "informed" contridictions. Somebody like sjoko2 could smoke my fanny on how all this actually works, but he tends not to get too deep into it here. Hopefully he will at least come along in this thread and say "Ed shut up, you don't know squat". Or "That's more or less the way it is folks" and possibly offer a few clarifications. I know I would like some clarifications about some of this stuff, but try getting ANY DSP manufacture to talk about it!!! LOL...fuck no they won't, because I suspect they don't want you to know how bad it really is, the cheap stuff (what most of us use...) and the good stuff (what the high dollar digital stuff uses).
Eddie
Almost EVERY application that works with 24 bit processing is 32 bit "float". A nice marketing ploy though.