What happens when you export audio at a higher bitrate?

Hi all:

I’ve always had the impression that my exported audio mixdowns sound just a little worse than the project playling live in Cubase. I’m not sure why this would be true, especially since many of my plugins offer the ability to use up to 8x oversampling on offline render and are using no oversampling at all in live mode.

So I never tried this before today, but I was doing a little cleanup on a 44.1khz project and I thought to myself, why not render it at 96kHz or at least 88.2, and see what happens. So I rendered it once at 44.1 and then a second time at 88.2 and a third time at 96.

First off, it works. I was a little surprised. Since the last time I thought about rendering at a different bitrate, I converted the whole project first and that was a disaster of ASIO overloading.

Second, the audio output was different in one way I could easily identify: the higher bitrate renders blew through the brickwall limite by a full dB. The 44.1 render peaked, as programmed at -0.3dB. The 88.2 overshot and peaked at +0.9dB. [Just found the intersample peak detection… all good here.] Aside from that, there may have been subtle differences in the audio, but they were pretty imperceptible.

Could someone tell me what the benefits/drawbacks are for rendering a project at 2x the project’s bitrate or higher?

If there are benefits, do I need to do anything special with the 88.2/32 file to convert it to 44.1/16 for CD or MP3?

Sometime ago I wrote a bit about sample rate and bit depth (you seldom talk about bit rate for PCM audio even though it’s perfectly well computable) here. Some additions though:

If you have a complete project with audio in a specific rate you get no real benefit from choosing a higher sample rate when exporting, you’ll end up with the same audio information plus additional samples interpolated to “fill in the blanks”.

The oversampling in your plugins does not mean that they suddenly give out a bunch of audio information in the domain above half of your project samplerate (in fact, that would give you aliasing distortion, quite the opposite from what you want) but is simply a matter of internal computational accuracy. A bit like the sample rate equivalent of 32 bit float.

/A

Read your post (and the one you linked to)… thanks.

So… since a lot of my music comes from Omnisphere, Trillian, and NI Komplete:

Question 1

Say my project sample rate is set to 44.1/32. I fill the project with just VSTi tracks with those instruments (no recorded audio). Then I do an export audio mixdown (offline render)–is there any sonic benefit of mixing down at 88.2/32?

Question 2 (only answer if the answer to question 1 was YES)

Now say I want to put that track on CD or MP3 (both of whch are 44.1/16). Is any sonic benefit still retained if I open this track in Sound Forge and save as MP3 or 44.1/16 WAV?

Answer one: No, if the project rate is 44.1 kHz and you just do the mixdown at a higher sample rate, then there you’ll end up with audio with no information above 22.05 kHz but with interpolated bulk samples added, i.e. extra large files with no benefit.

But if the original project sample rate is set higher from the start, and the VSTi:s you use support that rate then you might actually get some ultra sound frequency information. I say might, because even though they support it, that doesn’t automatically mean that they produce audio in that range. But, as I tried to point out in the linked post, even if you do come to the point where you have a recording containing ultra sound information, it does not automatically mean that it sounds any better or even is audible at all. It’s hard enough to construct audio electronics and speakers that can reproduce the audible range of sound, and you risk ending up with having all that high frequency modulating down into the audible range causing distortion for the end listener.

Answer two: If the source is 44.1 kHz, mixed down in a higher sample rate and then again converted to 44.1 kHz you’ll have no improvement in sound but risk a slight degradation from the steep filtering at 22.05 kHz necessary for avoiding aliasing distortion. If the filter is implemented correctly this degratation is of a purely technical nature though and should not be audible. Forget about MP3, one of the the first areas degraded in lossy coding is HF.

If the source has a high sample rate, then that’s an even bigger risk of getting sound degradation. In the example above (source is 44.1->mixdown in higher rate->down conversion to 44.1 kHz), there won’t be any actual audio contained in the ultra sound range. In a situation like 88.2->88.2->44.1 there’s the risk that some of the ultra sonic frequencies fold down into the audible range causing distortion. This is because every sample rate conversion is a tradeoff between the rejection of sound above the filter frequency and loss of that below, and there’s no such thing as a perfect filter. (but again, this is generally nothing more than a theoretical issue, most SRC:s produce aliasing well below the limit of 16 bit audio)

Hope you’ll get someting out of all this jargon, but the bottom line is:
Upsampling is pointless.
Using high sample rates throughout the chain, including end format, can be detectable with very high performance reproduction.
Downsampling should generally be avoided.

/A

Thanks! Makes sense.