24/96khz

[valhallasound]

A few of my algorithms actually downsample, so that they are running at 22/24 kHz. I do this to let the crappy linear interpolation artifacts SHINE.

MORE. NOISE. PLEASE.



In general, there are a lot of good reasons why 96 kHz might sound better than 48 kHz, none of which have to do with being able to hear frequencies above 20 kHz. A lot of signal processing algorithms have artifacts around Nyquist (I'm talking basic algorithms, like EQs and compressors). When Nyquist is at 24 kHz, this is pretty close to what people can hear, and the artifacts can extend into the audible range. Push Nyquist up to 48 kHz, and these artifacts are no longer audible.

Sean Costello

[@midnight]

That's a completely different context.

What happens to sawtooth waves in the upper register at 44.1khz is just plain nasty, and not in a good way.

Btw what samplerates are ValhallaDSP plugins compatible with?

[hibidy]

A few of my algorithms actually downsample, so that they are running at 22/24 kHz. I do this to let the crappy linear interpolation artifacts SHINE.

MORE. NOISE. PLEASE.



In general, there are a lot of good reasons why 96 kHz might sound better than 48 kHz, none of which have to do with being able to hear frequencies above 20 kHz. A lot of signal processing algorithms have artifacts around Nyquist (I'm talking basic algorithms, like EQs and compressors). When Nyquist is at 24 kHz, this is pretty close to what people can hear, and the artifacts can extend into the audible range. Push Nyquist up to 48 kHz, and these artifacts are no longer audible.

Sean Costello

That would have sounded better if it was done at 96k

[t3toooo]

so i'm not able to create a nice sweep @44....?

well,then....

[@midnight]

so i'm not able to create a nice sweep @44....?

well,then....

you can if it is oversampled internally, and you dont run any non-oversampled, distortion-generating plugins after it in the chain.

thats the thing if you are in 44.1 then you need to make sure all the key components are oversampled like the synth itself, any filter or distortion or compression plugins you might run after the synth

or you can just switch the host to 96khz, and you dont need to do any realtime oversampling/samplerate conversion, which itself can lead to artifacts

[james0tucson]

As of today I am now in the land of 96khz. Already everything is sounding more open and airy, crisp analog round fat gooey goodness.

Is that because the converters in your new expensive card are better than what's in your old cheap card?

[t3toooo]

In general, there are a lot of good reasons why 96 kHz might sound better than 48 kHz, none of which have to do with being able to hear frequencies above 20 kHz. A lot of signal processing algorithms have artifacts around Nyquist (I'm talking basic algorithms, like EQs and compressors). When Nyquist is at 24 kHz, this is pretty close to what people can hear, and the artifacts can extend into the audible range. Push Nyquist up to 48 kHz, and these artifacts are no longer audible.

Sean Costello

nevertheless,i'm surely not going 96k but this is a very interesting read,many thanks sean.


now the 96k police can further investigate.

[valhallasound]

That's a completely different context.

What happens to sawtooth waves in the upper register at 44.1khz is just plain nasty, and not in a good way.

Plug a Moog into your computer, and tell me how that sawtooth sounds.

If a software sawtooth sounds better at 96 kHz than at 48 kHz, then the sawtooth is not truly bandlimited. An additive synth, summing together all the sinusoids to make a sawtooth up to 24 kHz, will sound the same to humans whether it is played back at 48 kHz or 96 kHz.

Btw what samplerates are ValhallaDSP plugins compatible with?

44.1/48/88.2/96 kHz for sure. I don't have any way of testing above that. I also don't have any dolphins or bats working in QA.

Sean Costello

[valhallasound]

In general, there are a lot of good reasons why 96 kHz might sound better than 48 kHz, none of which have to do with being able to hear frequencies above 20 kHz. A lot of signal processing algorithms have artifacts around Nyquist (I'm talking basic algorithms, like EQs and compressors). When Nyquist is at 24 kHz, this is pretty close to what people can hear, and the artifacts can extend into the audible range. Push Nyquist up to 48 kHz, and these artifacts are no longer audible.

Sean Costello

nevertheless,i'm surely not going 96k but this is a very interesting read,many thanks sean.


now the 96k police can further investigate.

A simple mathematical example:

Take a sawtooth synth. Let's pretend that it is bandlimited to 20 kHz, and that the sampling rate is 48 kHz. Now run that through a linear compressor with a fast attack. This attack is essentially the same as amplitude modulating the sawtooth by a weird signal (the gain control), that has a fair amount of high energy due to the fast attack. The output of the compressor will have sidebands that are equal to the sum and difference of the sawtooth and the gain control signal. If the sawtooth has frequencies that go close to Nyquist, then the sum frequencies will be higher than Nyquist, and alias back into the audible range.

Now lets raise Nyquist to 48 kHz (96 kHz sampling rate). The sum frequencies of the compressor gain control signal and the sawtooth will be less than Nyquist, and won't alias at all.

It is interesting to think about how recording electronic music "in the box" is, in essence, a stress test for the various signal processing algorithms being used:

- Most "real world" signals are going to have far less harmonic energy than a sawtooth synth with the filter opened up the whole way. The closest non-electronic instrument would be a fuzz guitar, and this is usually being recorded through a speaker cabinet that is full of woofers (a 12" speaker doesn't have a lot of high frequencies). Cymbals generate high frequencies, but they tend to be inharmonic, so the aliasing is harder to hear. Transients are short, and a bit of aliasing can sound OK in some situations.

- Meanwhile, a virtual analog has a super buzzy waveform that ideally has harmonics all the way up to Nyquist, being processed by a filter that has embedded nonlinearities and a VCA with its own nonlinearities. Put this through some dynamics processors, and you've got a recipe for aliasing.

Sean Costello

[t3toooo]

very interesting again!

at the end i guess it's all about envelopes.a good mix @44khz can be as
good as a 96k mix.
yet you put em all those nasty aliasing's and distortions and mix them proper.

[lfm]

This is just silly. Higher samplerates can make a huge difference. It's not placebo at all,not even close. Do note i'm not saying it always will (depends on the plugins) or that it is the most effective way (it isn't) or that it makes sense to record at higher samplerates (it doesn't) ,i'm merely saying that "OMG PLACEBO" is way out of line.

One thing that seems to be forgotten in this discussion is - how much jitter is your machine causing at 96k compared to 48k?

If you are not getting the samples more accurate in timing when doubling the load on the machine you probably create more trouble than you solve.

You will get more artifacts if your machine and interface is not up to this load - and samples arrive a little now and then.

96k is not just a number - and bigger is better!!!! It's a really heavy load on your machine.

If to guess a little - usb interfaces will perform worse at 96k than 48k regarding jitter. A bit depending on the latency chosen, I guess.

A professional interface with wordclock though is probably alright - having this to sync it all. Besides - if it has a wordclock support says a lot by itself - highend vendor.

[whyterabbyt]

everytime you run a soft synth at a higher sampling rate, it will sound better

everytime you run any analog modeled plugin that has any nonlinearities (and that is most "analog modeled" plugins since like 2003) at a higher sampling rate, it will sound better

its been proven by the scientists already

as soon as you provide the scientific definition they're using for 'sounds better' i'll believe you. however, i'm pretty sure there's no actual scientific correlation between the conflation of 'higher frequency content' and 'sounds better' that you're making here.

as for 'analog modelled', im not entirely sure what aspect of that you're referring to, but if you're talking about higher modulation rates etc, then i'd be interested in comments from the respective DSP gurus round here as to whether higher sampling rates actually provide significant improvements in accuracy, result, comparison to oversampling the respective stages in the signal chain. and im aware that you've said that you consider oversampling has issues because of the downsampling stage, but Im also not yet convinced that this is results in provably 'worse' (ie not just a provably more band-limited) sound than a higher sample rate. comments from the respective DSP gurus would be welcome.

As for nonlinearities, I'd wonder if, for frequency content above the Nyquist limit for 44.1Khz/48Khz sample rates, most normal monitors and rooms contribute enough audible nonlinearities to render the point moot for most human's hearing.

In short, Im glad you're happy with your choice. It would be nice if you tempered the presentation of yourself as the only one in the room smart enough to have 'discovered' a new silver bullet, though.

[jupiter8]

This is just silly. Higher samplerates can make a huge difference. It's not placebo at all,not even close. Do note i'm not saying it always will (depends on the plugins) or that it is the most effective way (it isn't) or that it makes sense to record at higher samplerates (it doesn't) ,i'm merely saying that "OMG PLACEBO" is way out of line.

One thing that seems to be forgotten in this discussion is - how much jitter is your machine causing at 96k compared to 48k?

Could that be because jitter has nothing to do with this discussion ?

[t3toooo]

btw,this topic was well discussed in this thread one year ago.

http://www.kvraudio.com/forum/viewtopic ... sc&start=0



a lot of interesting suggestion and information.

myself is still not convinced to double the sample rate,simply because sooner or later it will come to the exactly same difficulty's in the mix,at least that's what i'm afraid of.


@hibidy you said that your distortion plugin sounds better at a higher sample rate?

i'm using a active pickup a sound devices 302 preamp and a rme aio,i tell you without a gate it is noisy as hexx but only because i compress the signal very hard.

i can dial my sound using for example guitar rig and a couple of plugins (eq or whatever) and i'm fully satisfied.


so whatever the choice is for somebody at the end you'll need to mix and to level and in my experiences this is the most difficult part,regardless of any rate.

last but not least,if this is aliasing,or i call it "blurring",i love it.
although distortion should not overlap in the overall mix but when aliasing is helping me than that's fine.

[lfm]

Could that be because jitter has nothing to do with this discussion ?

Yes, I guess that is why highend soundcards vendors carefully specify how much care they have taken to get low jitter(in picoseconds often).

The most extrem form of jitter would be when samples are missed alltogether - what we often call crackles and pops. A latency buffer is not filled until it is to be delivered.

But even arriving at an non-consistent timing will at some point affect how sound is perceived. A computer is a time sharing device - and not gapless. The more system stuggles the more likely to be dropouts.

Yes, offline rendering will not be affected - but we talk about quality while mixing and until stuff has been recorded ITB all this matters.

All A/D and D/A conversion rely on a steady stream of samples. So all recording may be affected.

At 96k every sample should be 10.416us apart. If system is struggling and samples are jumping maybe between 10.000us-11.000us apart it will not reproduce original sound and artifacts are there.

Latency buffers are there - and you will probably have to go up a little bit more than doubling according to sample rate increase. If running fine at 64 samples 48k, you might have to go to 256 samples running 96k - if to run the same amount of tracks and plugins as before rendering down to stems or freezing tracks.

And further that professional soundcards support wordclock to carefully sync all parts in studio with minimum jitter. A highly stabilized clock.

It has everything to do with audio quality and artifacts(and crackles and pops as well). Exactly how much is perceivable is hard to say. The developers here can probably enlighten us a bit.

But as I said - it's forgotten in this discussion that the much heavier load on computer with 96k affects many things.

And all ADAT ins/outs are reduced to half because of the heavier load on interface. Only 4 channels/ADAT connection at 96k.

Most daws don't handle that so well, so if jumping between projects where some are 96k and some 48k your in for some extra configuring in between.

A soundcard that does not support wordclock is consumer level products is my assumption. And to even bother running 96k is most likely a waste - at least not to it's potential.

Just my take on 96k. Maybe my next daw in 5 years.