Please note: I wrote the following to clear myself on this issue. I don't mean to sound like I'm accusing anyone else of being wrong. I'm pretty new to all of this crap myself. BTW, I just noticed that we seem to have a lot of engineers around here....makes me feel like less of a geek
----
Whoa, everybody but Drstawl is confusing me here.
Why do you keep using time in your explanations of bit depth? The number of "samples" per second should be dictated by the sampling RATE. In most cases 44.1khz. That's where the strobe light and image resolution analogies fit in.
When recording at 44khz, your guitar sound, for example, is being sampled 44,000 times every second. When you playback your recording at 44khz, you're hearing 44,000 samples per second which is pretty close to continuous. In the real world, you hear live sound as a continuous function. In the computer world, you're dealing with discrete functions. Think movies here. In real life you see images as light is constantly hitting your eye. At the movies, the light is only smacking your eye 24 times per second which can be thought of as a "sample rate" rate of 24hz.
Now let's talk bit depth in terms of the image analogy that was posted. Increasing the NUMBER OF COLORS can result in a more accurate picture (talking real-life pictures, not clipart or anything). A pixel in a 16bit image will be one of 65,000 colors. A pixel in a 4bit image will be one of 16 possible colors. This is analogous to the bit depth of your sound. Now, this has NOTHING at all to do with image RESOLUTION which is just as important as bit depth. Image resolution is simply the number of pixels across by the number of pixels down and is analogeous to SAMPLE RATE.
Now let's talk music. Say you record with a sampling rate of 2hz. That's two samples per second. yeesh. Now let's say that you record a one second bend on your guitar. What are you going to hear? A little blip at the start of your bend...then silence...then a little blip at the end of your bend. Increase the sampling rate to 3. You hear a blip at the beginning, middle, and end. The more you increase the sampling rate, the less silence and the more guitar...which is a good thing!
Bit depth in terms of music: think back to the color analogy. If you take a picture of your goldfish and scan it in as 4 bit color, then you only 16 possible colors to represent your goldfish. In real life when you're staring at your goldfish, there are millions of different colors hitting your eye that make up the overall "color" of the goldfish. When you try to represent all of those colors with only 16 variations, then you're goldfish is going to end up looking like a cartoon character. The same thing happens with your music. Higher bit depth helps to capture more of the subtlety of your sound. It will sound more real...more alive.
Need to think in terms of math? Sound is a wave form hitting your ear drum. A nice continuous function that moves "up and down" with time. Sampling that wave would be a process of cutting it up on the x axis. Let's say that you sample that wave 1000 times every second. That means that you take a sample at time 0.000 seconds, time 0.001 seconds, time 0.002 seconds, time 0.003 seconds, etc all the way up to 0.999 seconds. Great, that's a lot of samples. But what values did you aquire for each sample? Now we come to the Y axis (start thinking bit depth!). Let's say that our accuracy for measuring the wave is 1. Our data (don't worry about what it IS) might look something like: 1, 1, 2, 1, 2, 3, 1....etc etc up to 1000 values. That might be just fine according depending on your actual needs, but let's say that it's not. Increase your measuring accuracy to 0.1. The data might now look like 1.4, 1.9, 2.0, 1.8, 2.4, 3.0, 1.9... See how things start to clear up? Now increase the accuracy to 0.01. The data might look like 1.92, 2.01, 1.83, 2.48, 3.01, 1.92... Hmm, increasing our accuracy in this case didn't seem to help much at all because we still have some pretty big gaps to fill in! How do you get rid of the gaps? MORE SAMPLING! Sample that sucker 2000 times per second. The data becomes 1.92, 1.95, 2.01, 1.94, 1.83, 2.02, 2.48, 2.79, 3.01, 2.52, 1.92... Aha, it's starting to smooth out a little bit...not so many big jumps. Now sample the sucker 44,000 times per second. Holy smacks, the big jumps will disappear after which the only way to become even more accurate is to INCREASE THE ACCURACY. See how the two are intertwined and equally important? Accuracy in this case, if you haven't figured it out, would be BIT DEPTH. The number of times you look at a wave over a fixed period of time is your SAMPLE RATE.
To sum up this big giant mess of a post: Sample Rate is the number of times you take a "measurement" of your sound every second. Bit Depth is how accurate each "measurement" actually is.
Bit depth without a high sample rate is useless. High sample rates without any bit depth is just as useless.
What's TRUELY important is how well you are able to recreate the original sound. Your computer isn't going to store an infinite number of samples at an infinite bit depth. But at some point it's just not important anymore because your ear, like any tool, is limited in accuracy! Just like the movies fool your eyes into seeing Pamela Anderson's boobies bouncing around in front of you, your CD player fools your ear into thinking that Tad is jamming in the corner of your room.
So this of course beggs the question: HOW MUCH DOES IT TAKE TO FOOL THE HUMAN EAR? You could make a soundcard that would sample 100,000,000 times every second at 1024bits. Holy crap that'd be the best soundcard in the whole world right? Well, I suppose, but it probably wouldn't sound any better than the current top of the line 96khz 24 bit soundcards or whatever.
Everything in your recording lineup is a tool and tools are only as accurate as humans can make them. The only thing that's actually REAL is that little vibrating string on your guitar. Cool.
Welp, I just made all that crap up. Please let me know if it's wrong.
Slackmaster 2000
[This message has been edited by Slackmaster2K (edited 11-18-1999).]