The Song Fight Bulletin Board

Sven wrote:Always work with 48kHz until you encode to final mp3.

blue wrote:

Sven wrote:Always work with 48kHz until you encode to final mp3.

that's not really good advice for people who are doing a songfight collab.

Deshead wrote:Read Tweak's article about bit rate as background. When I say "bit rate," I'm combining the concepts of "bit depth" and "sample rate."

BIT DEPTH

The number of bits used to capture a signal determines the dynamic range, the maximum difference between the softest and loudest sounds in the signal.

The math tells usthat 16 bits gives 96db of dynamic range, where 24 bits gives 144db. Here are some other numbers to consider when thinking about dynamic range:

The noise floor in your studio is probably 40dB SPL. If you mixed in a library, you'd be down around 30dB SPL. If you mix at 85dB (the sweet spot in human hearing) you're still 45-55dB above the noise floor of your ideal listening environment. You'd need to track things 15 times quieter (that is, at -60dB in a -15dB RMS mix) to bring the 16-bit noise floor above the masking noise in your environment.

But ok, fair enough, it coould happen (<ahem> ) So there's a case, at least for gain-staging amateurs, to use the extra headroom provided by 24 bits.

Well, if some is good, then more is better, right? Why not go all the way to 32 bits, or higher?!

144dB, the dynamic range of a 24 bit recording, is outside the realm of "hearing" differences, and into "feeling" them. If the quietest sound you capture is a whisper, then the loudest sound you can capture is anywhere from 5 to 8 times louder than a jet engine. Either way, enough to instantly perforate an eardrum.

Most condenser mics fall apart above 120dB. Even the dynamic mics I could find specs on exhibit distortion above 140dB. And while I suspect you could put an SM57 inside a rocket engine at takeoff, here's the real kicker: I am unable to find speakers capable of an SPL beyond 131dB. There's not a listening system on the planet capable of reproducing the full dynamic range of a 24-bit recording.

One last note on bits: The 32 bit or 64 bit processing in your DAW's marketing material refers to internal signal processing, changes that happen to the signal after it's been captured. Digital signal processing involves math, and the more bits a computer uses to store numbers, the more accurate the math is. (Remember learning how to multiply fractions with a calculator?)

So there's a good case for using more bits to work with the sound once you've captured it. And if you were limited to one bit depth throughout the entire process, from recording to mixing to mastering, then you could strongly argue that starting with the highest bit depth possible is a good idea. But that's not how it works. Your DAW doesn't care how many bits you used to capture the input signals. It's going to do all its math using 32- or 64-bit floating point numbers.

In short, then, if you're smart about gain staging, you can probably record in 16-bits, and no one will be the wiser. But even if you want the extra headroom, there's no reason to ever capture sound using more than 24 bits.


SAMPLE RATE

In a digital system, the frequency at which you sample an input signal determines the highest frequency you can reproduce on playback.

Nyquist figured out most of the head-spinning math. He shows us that sampling at 44.1KHz, we'll capture signals up to 22Khz, 2Khz beyond the accepted limit of human hearing. Audiophiles claim that humans are sensitive to frequencies higher than that, so we miss part of the experience from any recording that lacks those frequencies (especially CDs.)

Recent studies suggest that we are, in fact, sensitive to hypersonic signals. But my hero Dave Moulton spells out why this is irrelevant to audio engineering: Accuracy is a myth.

It's not possible to make a recording that faithfully, perfectly, recreates a sound source. We all know this. A mix sounds different on every system, and the frequency response of human hearing is incredibly volume-sensitive. Those are simply facts of life for us mix engineers. So instead of "perfect accuracy," we strive for "transparency" in mixes. That is, mixes that translate well from system to system, never perfect, but always good. More simply put, our art is imperfect.

The audiophile completely overlooks this. It's an act of hubris to assume that a $5,000 amp and $10,000 speakers will give a more accurate or perfect sound than what the mix engineer intended. And this leads to an important point: unless you're dealing with a single-source recording through transparent equipment (as did the study linked above,) everything you hear is the result of a mixing decision.

There are two reasons I can think of that one might decide to capture sound at a sample rate beyond 44.1KHz:

The first is that HFC allows for greater accuracy. But as I describe above, accuracy is meaningless in recording. Once a listener touches the volume knob, any notion of accuracy vanishes.

The second reason then follows: Even if it doesn't make things more "accurate," it can't hurt. Most systems won't reproduce the sound anyway, but the ones that can will enjoy an improved experience.

When I hear this from mix engineers, it troubles me.

The argument that 44.1KHz recordings are less enjoyable because high frequencies are missing implies a few assumptions:
- That the recording gear is sensitive to the desirable high frequency content (HFC.)
- That the montioring environment allows the mix engineer to make decisions about that HFC.
- That the mix engineer is sensitive to HFC in an objective way.

These conditions can certainly be met, and I'm sure there are mixing engineers in the world who satisfy all 3 assumptions. But before deciding to use a high sample rate, you have to ask yourself honestly if you meet the conditions. In fact, unless you have the equipment to accurately capture and guage HFC (most pro-sumer mics and speakers start to roll off around 17KHz,) and believe you can objectively mix the signal, the notion that "adding it can't hurt" is antithetical to everything a good mix engineer stands for. Remember, transparent mixing depends on making decisions that improve the mix. Every change you make should improve the final sound, or else it's simply not needed. Adding HFC "just because" is equivalent to slapping a compressor on every track for the hell of it.

We all know not to do this, and the same applies to the decision about which sample rate to use. Unless your equipment and skills are up to the task, tracking at 48Khz or 96KHz is actually DAMAGING the sound ...



As all this applies to Gert, I hope my position is obvious: Since none of us have the equipment or skill to properly address HFC in our mixes, we should use a bit rate of 24/44.

And if the day comes that a big label acquires the skill and equipment for us, we'll have much better things to worry about than re-tracking everything at 48Khz.