Sample Rate Change On Render

I just found out something rather shocking reading the manual.

Normally I work on 88.2kHz, this way i get hi resolution samples of my hardware synths. When i play a sample down an octave it should have a rate of 44.1kHz.

But when i render a song i used to just render to 44.1kHz. I was browsing the manual and read the rendering section more or less by accident. It said this influenced the final mix cause some plugs handle this differently.

Can somebody tell me a bit more about this? What plugins? And what makes them sound different? Are they bypassed?
I also want to suggest a popup warning in renoise before rendering to a different sample rate.

I already thought i heard a difference between the bounce and the project but blamed my ears/ interpolation type.

88Khz consists of twice the data that 44Khz delivers, so when you downgrade from 88Khz to 44Khz, you miss out on 50% of the data. It depends on how dynamic the sound is though, the more symmetry a sound has, the less likely you will notice. The lower the frequency range a sound falls in, the less likely you will notice.
But high frequency ranged audio or noise will definately have noticable effects. This is a common thing with digital audio though, not something Renoise specific so this applies to all audio processing and editing software, but we simply decided to put the warning in the manual anyway because not everybody is aware of that.
But the majority of people do know this fact so a warning dialog didn’t made it because of that.

Plugins that are affected means:plugins have their own internal maximum of sample-frequencies they can support. A good plugin can support up to 192Khz, the more low-end range plugins go as far as 44Khz and they supply funny output when you go beyond that scope. But you will notice this direct as the sound comes out twice as fast or twice as slow when a plugin is processing it.

Doesn’t really matter though since you’re not going to hear that data anyway.

Not with any decent modern DAC.

When you’re working with a track especially when you’re doing mixing or mastering it’s good to have the samplerate as high as you can, but exporting to anything higher 44,1khz doesn’t really make any sense (assuming you’re using the sinc interpolation), unless you’re planning to further edit the export with another program. But if there are problems like the OP described just export is at the samplerate you used while making the track and the downsample it afterwards.

Downsampling is a crummy mathematical progress, a synthetic way of getting a compromised result. It’s not flawless.

Better to work in as highest rate you possibly can and do one of the follow: 1. Give the higher rate render of the mix to a mastering engineer so they can properly handle making a 44.1khz result. 2. Play out a higher rate render through a good DAC at 24bit (dithered from 32bitfloat), run through whatever outboard gear applicable, and have a seperate ADC capturing the result at 44.1khz (capturing a dithered result of 16bit, real time or not, up to you). Assuming your clock quality is good, jitter low, then you’ll get a much more organic result compared to the synthetic process of downsampling.

A higher sample rate has implications right across the frequency range, not just the tops. It is not strictly matter of whether or not you will hear more detail in the tops, but more generally one of more resolution for all frequency types. Mids will sound richer. Bass will sound more dimensional. So the inverse applies to limiting the sample rate. You hear the result either way, and there is a difference, despite whether or not you’ve cultivated your hearing to an extent to be able to confidently hear character of the difference.

What are the implications for getting it wrong or right? The longevity of your work.

I know converting sample rate has its effect, but this is the line in the manual:

Many DSPs may sound slightly different at other rates, so changing the rate may result in a slightly different sound from what you expect.

As i read this now, its not only about rendering but also changing the complete project setup to a different sample rate. Does a project made in 88.2kHz sounds different when i give it to somebody who works in 44.1?

On the sample convert side of things… Isn’t a tracker like renoise all about converting sample rates? It does it all the time playing samples at different pitch (sample rates). Why would i have to put so much effort in bouncing a master at a different sample rate if the whole song is acctually mostly at a different sample rate… But this another discussion as it’s not about vst behaving different at different sample rates…

Yes. It will sound worse.

Suggestion. Bounce to disk at the original sample rate, then use a high quality sample rate converter, like SoX, to downsample to 44.1khz. You really should take great care in how you down sample.

See here:
http://src.infinitewave.ca/

Most sample rate conversion algorithms are rather dirty, unfortunately Renoise isn’t the best (sorry).

The best tool for Freq conversion (and its FREE) is Voxengos r8brain. Check.

r8brain is on the list, and it checks out. According to http://src.infinitewave.ca/ the Minimum Phase r8brain is just as good as the Minimum Phase SoX implementation per the sweep images. This is pretty fantastic, considering they are both free.

Minimum phase is not free, I think, since it has a Pro thing to it. The Linear phase performs excellent in all tests on par with iZotope’s sample rate converters, however the free version treats phase as if it is non-existent. It is a different question, whether one needs it in most cases (mp3 format seems neglect the phase too and no-one has been complaining so far) but that still keeps the question of representation fidelity open.

my bit))
why more khz is better? example: 16khz square wave (you can heart it, my “fantom” limit is 18khz 0db)
384khz - 24 (384/16) samples per wave. square restore is perfect
192khz - 12 samples per wave. almost perfect
96khz - 6 samples per wave. after interpolation looks like sin+square. good result
48khz - 3 samples per wave. after interpolation looks more like sin than square. bad!
the same for saw and more complex waveforms
pure sin even at 44.1 is close to perfect. but electronic music is not a sin wave! :D

Nice work with the pictures to explain! (Y)

Unfortunately, a square playing at 16kHz would sound exactly like a sine playing at 16kHz, given the first harmonic for a perfect square at 16kHz would be 48kHz. Only aliasing will give a square wave audible harmonics for a fundamental playing that high.

Higher sample rates will help with the samples you use, particularly when they’re played back a much lower frequencies, but as far as overall playback or mix down… There’s still much ongoing debate on whether or not there’s any real audible difference or if it’s really psychological, i.e. It’s a higher sample rate so it must sound better.

I’ve given up on working with anything above 44.1kHz for mixdowns. The only difference I’ve ever noticed for higher sample rates was that my hard drive was filling up faster. Sample size is far more important there, particularly if the mixdown is source for mastering. 24-bit if you have to, but IEEE 32-bit float is better.

Exactly! A square wave playing at 16kHz will sound exactly like a sine wave playing at 16kHz to any human as we can not hear the frequency of the first harmonic anyway.

But:

What a lot of people miss is that the real different comes in representation of levels and transients, rather than pure frequency response.

I’m not going to try and make any mock-up pictures, if my words confuse you I’m sorry. If people need more expansion I can try and dig out an article or something later though, hopefully illustrated ;)

Let us assume we have a choice of sample rates of 48kHz, 96kHz and 192kHz (chosen as they are common and all easy multiples of each other.)

If we have a sine wave of exactly 24kHz (which we will ignore is outside of the generally accepted range of hearing) then the rate of 48kHz will produce two samples per cycle, giving perfect reproduction of frequency on decoding. But let us think about the phase/timing of this waveform. You are only going to get correct reproduction of the amplitude of this wave if the sample points fall exactly on the peaks, which is very rarely going to be the case. Shift the phase just a little bit and the sampled level will be lower, then at certain phases you will be sampling at the zero crossing and actually miss the wave entirely.

Now if it wasn’t exactly 24khz but rather something like 23.5kHz then the sample points compared to cycle position change over time. You record a steady, constant tone and play it back and it will get louder and quieter cyclically.

I don’t think I really need to explain why having a higher sample rate improves matters there…

How noticeable it is in a complex sound is open to argument. Especially when you take into account that the very top end is only used for harmonics of other sounds generally and a lot of people beyond their teens have to at least some extent degraded the upper end of their hearing.