Upsampling Sounds Prior to Processing

Question

Let's say you've got a batch of raw sounds you're working on for your library, making versions of them that are more 'production ready.'

Let's say that these sounds are 44.1 KHz / 16 bit.

I find myself doing this quite a bit. Running raw sounds through distortion plugins, outboard guitar amps, guitar pedals, playing the sounds through a speaker and putting up a mic in the room, running them through reverb plugins, processing them through outboard 'synths', etc... I find that having different versions of sounds around, processed in interesting ways, can make things easier and more interesting when I'm putting together a full piece later. So all these processed sounds serve as 'building blocks' for full sounds.

Moving on, here's the question:

Is it worth upsampling these sounds (necessary when importing them into a DAW session to match the session's sample rate whilst playing back at normal speed...or recording the results to a higher sample rate if playback and recording happen on two different devices) to, say, 96 or 192 KHz (and a higher bit depth like 24 as well)?

Is it safe to assume that harmonic distortion will, for example, create harmonics of even a 44.1K file that would require a higher sample rate to capture those harmonics?

And what if they're sounds that will be, for example, pitched down, layered with something else, and eventually pitched back up through some other process down the road when making more building blocks?

I don't necessarily want to expand the size, in MB or GB, of my library simply because I think it's cool to have a bunch of 192's in it for the sake of having higher sample rate files only. Is there a benefit beyond that ridiculously pointless reason?

I've experimented with pitching a 16 bit / 44.1K file up and back down and up and back down over and over again with, while interesting (and possibly usable for something interesting), absolutely dismal results (which, I would expect).

Thoughts?

fonte

Answer 1

Stavrosound already did an excellent job covering conversion of sample rate (followed up by a nice addendum by Dave), so I'll cover the bit rate part of your question.

None of the DAW's we use work natively in 16-bit. Before people starting crying foul, I'm not referring to the actual file's sample rate, but I mean within their computational engines. They all work in higher "floating bit" rates once the file is actually pulled into the session. Pro Tools HD, for instance, uses 64-bit floating point architecture within its engine. It's important to note that I'm also not referring to the host processing system's specs. As far as CPU's go, Pro Tools doesn't yet make use of the full power available in 64-bit processors (and the non-hd version uses only 32-bit floating point). That doesn't mean that it can't use a 32-bit processor to calculate products and sums of 64-bit words, because that's exactly what it does. So, all of your volume automations, fades and plug-in processing is occurring natively at bit rates much higher than 24-bit. It's only on output that it gets truncated/converted back to 16 or 24 bit.

That means that increasing a file's bit rate prior to bringing it into a session is relatively pointless. If you're going to create derivative sounds using lower bit rates, you can output at 24-bit and still retain any benefits you gained in the audio engine. If you start with a 16-bit file and output a final 16-bit and 24-bit one, the 24-bit will have more dynamic range information than the 16-bit. For this reason, for any new sounds you create (regardless of the source files' bit-rate), I suggest always outputting at 24-bit.

It also goes without saying that you should always record at 24-bit. ;)

Answer 2

Well, this question depends on a few things in my opinion. First, if you run a assets manager like Soundminer, Basehead, or Netmix, it's going to convert the source audio to match your session when you spot a file in. With Soundminer HD Pro's 64-bit iZotope SRC (now multi-threaded), its very quick that it's almost an invisible process on all but very long sounds. This means that your library can be stored in a garden variety of 44.1-192k files, and the program convert them on-the-spot to match your project.

By upsampling, you will increase resolution of the plotted voltage points so it will be more malleable, however (and this is a big HOWEVER), it will not create new audio content - it will just be redundant/interpolated points of what already exists.

An example of this is with the Nyquist Limit, where a signal's max frequency is half of that of the sample rate. For 48,000 Hz file, that means 24,000 Hz is your theoretical maximum frequency. If you then play this sound at half speed, it cuts the signals frequency in half to the max being only 12,000 Hz - meaning now your super high frequency crispness is beginning to deteriorate. As you go even slower, you loose the top even more and it starts sounding like a washy mp3. Whereas if you have a 96k recording (assuming both the mic and A/D and D/A is capable of 96k throughput response), your max frequency content is 48,000 Hz. So, when you cut that in half for a half-speed playback your max frequency is 24,000Hz - this is still high enough that it provides full sonic clarity, and might even have some cool sound content up there that used to be inaudible. Frank Bry mentioned this frequency shifting in a recent post about recording bats and they were analyzed to make noise at 39,600 Hz, but it was only discernible once he shifted the sound down by 2 octaves. In a true 44.1 or 48k recording, this sound wouldn't even be captured, and when you cut the sound speed in half, you wouldn't get close to finding anything up there but empty bits or white noise

So I guess what I'm getting at here is that for upsampling: yes, you will have more math for number crunching (e.g. wasted hard drive space). But, you will not get the extra fidelity that true 96k+ recording has when you are prepared to run it through the wringer.

These are my 2 cents, and somebody please correct me if I'm wrong somewhere.

Answer 3

Stavrosound is absolutely correct if all you're doing is upsampling the sounds in your computer. But, if you're running them through offboard gear first, then definitely record them at the higher sample rate. Guitar amps, pedals, and other analog gear doesn't think in terms of sample rate, so the results coming from your offboard processing very well have nuances that the extra headroom and clarity will afford you.