Hey,
I tried that Nord G2 demo by clavia the other day, and on the setup screen I had the option to choose A sample rate. the minimum was 44100 but the recommended frequency was alot higher, like 9k or something like that.
It got me thinking: does it really affect the sound quality of VSTIs if you crank up the sample rate of your host program? (mt card is capable of higher frequencies) I thought it was only relevant for recording...

Oh, man. This thread's gonna go mental!

At the risk of sounding banal, I reckon it's different strokes for different folks here. I've never really felt the need to go higher than 44100 myself, though that's mainly because I prefer to have the extra processing power available rather than (what is to my ears) a barely perceptible increase in sound quality. Others will (and are going to ) tell you differently though.

I think that greater bit-depth [24-bit or 32-bit] is far more important than higher sample-rate but if your PC can handle both then why not work in the highest quality you can? That said, I always record/render at 44.1/32-bit.

BONES wrote:

I think that greater bit-depth [24-bit or 32-bit] is far more important than higher sample-rate...

Agreed!

Disagree here. Higher sample rate will give more accurate recreations, and on some VST's, yes it does make the sound a bit smoother. You're playing connect the dots with 96,000 points or 44,100 points over a one second segment. You tell me which would be more accurate and smoother? Also, 96k is going to prevent some aliasing too, at least for freqeuncies up to 48kHz instead of only 22.1kHz.

Devon

Quite a few synths do profit from an increase in sampling rate as its just like oversampling. I.e. you can get rid of aliasing this way (e.g. try vanguard at 44khz and then at 96).

Markus

even to my mono hearing 96000 sounds a HELL of alot better. it's just that current cpu's cant keep up with a heavy vsti project of 96k. no way.

Well, could the people who did feel a different be more precise? is it a smoother filter? different oscillator sound? I know it sounds better in theory, but DOES IS REALLY SOUND BETTER TO YOU?
How about after your dither it to a 44100 wave?
And I'm talking about strictly VSTIs, not audio.

Download the vanguard demo, try it at 44khz first, then run it at 96khz.

I tend to work at 44k, since 96k just taxes my computer too much.

However, I then render at 96k and then downsample and dither in a separate step.

higher samplerates will always sound a bit better, but if a plugin is well programmed it should be a very minimal increase in quality (due to selective oversampling inside the unit).

Kuniklo: Could you elaborate on the logic behind that procedure?

UvRayz wrote:

Kuniklo: Could you elaborate on the logic behind that procedure?

Work at 44.1k to keep the load on the computer to a minimum, bump it to 96k and render to get the best sound out of the plugs.

If you're using recorded audio rather than everything being generated on the fly, I suspect it would be (perhaps marginally) better to render at 88.2k instead of 96k. The sample rate conversion process, which you would be going through twice assuming you will be converting the final product back to 44.1k, is more straightforward in that case.

DevonB wrote:

You're playing connect the dots with 96,000 points or 44,100 points over a one second segment.

No, you're not playing connect the dots; yes, in SOME VSTi there's a huge difference in sound between different sample rates (in my experience Lounge Lizard, Powercore 01, Superwave synths etc.; NOT in Pro53, B4 and some others), but it hasn't anything to do with the dots. Sorry, I'm not the best to explain what's happening exactly.

Below the nyquist frequency (Sample rate / 2) there is no theoretical advantage to the apparently higher resolution of 96k. Any benefit lies in the lower order (potentially more linear) anti-alias filter that can be used with the higher rate.
Whether this can be heard is a moot point.

nuffink wrote:

Below the nyquist frequency (Sample rate / 2) there is no theoretical advantage to the apparently higher resolution of 96k. Any benefit lies in the lower order (potentially more linear) anti-alias filter that can be used with the higher rate. Whether this can be heard is a moot point.

Some vsts sound very noticeably better at higher sampling rates. The difference can be very obvious.

UvRayz wrote:

Kuniklo: Could you elaborate on the logic behind that procedure?

It's pretty straightforward. Some synths either sound duller at 44k or have very audible aliasing that goes away at higher sampling rates. So, you render at 96k/24bit to get the best possible sound and fewest artifacts. The resulting file can then be downsampled and dithered to the highest possible 44k/16 bit quality. If you render at 44k there's really nothing you're going to be able to do to remove the resulting artifacts.

I think this is the way a lot of pro studios work. They record at the highest practical resolution and then downsample/dither to CD quality for release.

kuniklo wrote:

It's pretty straightforward. Some synths either sound duller at 44k or have very audible aliasing that goes away at higher sampling rates. So, you render at 96k/24bit to get the best possible sound and fewest artifacts. The resulting file can then be downsampled and dithered to the highest possible 44k/16 bit quality. If you render at 44k there's really nothing you're going to be able to do to remove the resulting artifacts. I think this is the way a lot of pro studios work. They record at the highest practical resolution and then downsample/dither to CD quality for release.

So how does this help with aliasing? I understand that you don't have aliasing with 96k, but when you downsample to 44k doesn't it result in aliasing still?

Not too long ago, I read from someone here that higher sample rates=lower latency. Is this true, it doesn't seem right? (wish I had asked the person when I read it)

Despite claims either way, some synths are said to sound better rendered as 96khz. Whether there is any point in recording at 96, there is probably a strong basis for this, and it should be easy to do a blind test of the claim.

But the community as a whole seems satisfied in dismissing any benefits to either wide bit depth or high sampling rates, and this ends up raising the barrier for people who are trying to do things with sound other than music. In particular, it is obnoxiously difficult to find amplifiers that can work with sonar frequencies, and nearly impossible to locate speaker drivers that can accurately reproduce very high frequencies.

This makes life difficult for researchers in the biological and physical sciences. I hope you don't think all university researchers have an unlimited budget, or any budget. Oh well, gotta get on my plane.

kuniklo wrote:

Some vsts sound very noticeably better at higher sampling rates. The difference can be very obvious.

Which does not neccessarily say anything about whether the higher sampling rate sounds "better", just that the particular VST sounds better at that sampling rate.

I used to do a lot of reading on this subject, of posts and papers written by people far more knowledgable than I. What I basically got from all of it was that 44.1 is not ideal, 96 is probably a bit more than is useful but fine to work with, and 192 is very excessive and can be worse when doing AD/DA. The simplied and rephrased explanation of this being that you trade speed for accuracy, 192 is too fast, you lose accuracy, therefore losing quality. I've frequently seen "60-80" listed as being (probably) optimal for music use, with 88.2 and 96 being cited as the most practical rates to achieve the best quality.

I'm just regurgitating the recommendations that came out of the arguments, though, and won't argue the theory behind it all. I think that's best left to people with degrees and years of professional experience in this stuff.

edit- Freeztar: latency is generally measured in samples, 256 samples is a longer period of time at 44.1 (44100 samples per second) that 88.2 (88200, of course), so yes; assuming you can run at the same latency settings.

freeztar wrote:

So how does this help with aliasing? I understand that you don't have aliasing with 96k, but when you downsample to 44k doesn't it result in aliasing still?

It depends on how you downsample. I'd imagine most good downsampling utilities will put some kind of steep filter around the nyquist frequency to eliminate frequencies that could alias down into the audible range.

contrast wrote:

Which does not neccessarily say anything about whether the higher sampling rate sounds "better", just that the particular VST sounds better at that sampling rate. I used to do a lot of reading on this subject, of posts and papers written by people far more knowledgable than I. What I basically got from all of it was that 44.1 is not ideal, 96 is probably a bit more than is useful but fine to work with, and 192 is very excessive and can be worse when doing AD/DA.

Right. I think it's important to distinguish here between the behaviour of vsts at various sampling rates and theoretical considerations. The fact that some vsts sound much better at higher sampling rates doesn't mean that we hear anything above 22khz at all. I'd probably still record at a higher resolution, just because I like to have the extra headroom for manipulation, but I suspect that well mastered audio at 44k final resolution should be good enough for just about anything.

kuniklo wrote:

freeztar wrote: So how does this help with aliasing? I understand that you don't have aliasing with 96k, but when you downsample to 44k doesn't it result in aliasing still? It depends on how you downsample. I'd imagine most good downsampling utilities will put some kind of steep filter around the nyquist frequency to eliminate frequencies that could alias down into the audible range.

Aha, that makes sense now.
Thanks.

contrast wrote:

edit- Freeztar: latency is generally measured in samples, 256 samples is a longer period of time at 44.1 (44100 samples per second) that 88.2 (88200, of course), so yes; assuming you can run at the same latency settings.

Ok, but I'm still uncertain about something here. Sure, you get more samples in a shorter amount of time, but it is still the same audio. It seems to me that the extra processor time of computatiing the additional samples would offset any benefits you would gain from extra samples.
In other words, the time domain of heard audio is the same (latency), but the sample content (processor usage) is higher.

The processor usage will be higher at higher sample rates, yes. That only means that you will be able to do less before running into CPU issues; crackling noises, etc, I'm sure you've asked a bit too much of your CPU at some point.

You may even then decide to increase the latency of your soundcard or plugins (where possible) to help free up the CPU a bit.

But if you can or choose to run at the same latency settings (measured in samples) the latency in terms of actual time will go down as the sampling rate goes up.

contrast wrote:

But if you can or choose to run at the same latency settings (measured in samples) the latency in terms of actual time will go down as the sampling rate goes up.

Will you please elaborate on this? First of all, the latency settings I have access to are measured in milliseconds (ie Tracktion, My ASIO Aardvark card is at 4.3ms). Secondly, how (exactly) do more samples in the same amount of time=lower latency. I know I'm being thick, but this information could be very useful to my ignorance.

Well, what is confusing you here is that sometimes latency is measured in samples, and sometimes it is measured in milliseconds.

If the latency settings are measured in MS, then that's the latency you get. 4.2 MS is 4.2 MS. The software takes care of figuring out how many samples this works out to on its own.

In the case that latency settings are measured in samples, you only need to understand that it takes more time to play X number of samples at 44.1KHz than it does at 88.2KHz. Twice as long in fact, because half as many samples are played every second.

at 44100, you've got 100 samples in a 441Hz wavecycle.. 2 octaves higher, still well within musical range, you've got 25 samples per wavecycle..

12.5 samples per 180 degrees..

go.. 2 octaves higher.. 3.125 samples per 180 duty cycle..

that's not especially accurate, tho i don't worry about it myself since i use audio semantically.

That's not a good way to think about it. The number of samples per cycle is irrelevant. In theory, at 44.1, any frequency below 22050 will be reproduced perfectly. A sine wave at 441Hz is not somehow better or higher-res than a sine wave at 882Hz, they will both come out as perfect sines.

In practice, how well they come out depends on how well your AD/DA works, but that's a separate issue.

This begs the question of what, exactly, is being lost; there are fewer samples per cycle, that means less information per cycle, so logically we are losing something. What is being lost is, and is only, the portion of the wave that contains frequencies above 22050.

contrast wrote:

In the case that latency settings are measured in samples, you only need to understand that it takes more time to play X number of samples at 44.1KHz than it does at 88.2KHz. Twice as long in fact, because half as many samples are played every second.

Ok, Still, however you measure it, it should equal the same amount of audio time (bpm) in the end. Sure you can play the same amount of samples in 44k as you do in half the time at 88k. But what hosts/vsts work this way? Also, samples are not strictly time related. Latency is however (cpu load).

Visually:

1---2---3---4 (44k)
1-2-4-5-6-7-8 (88k)

How does the latency change here? (assuming that latency is a measure of the time difference between input and output)


What you are seemingly saying (visually):

1---2---3---4--- (44k)
1-2-3-4- (88k)

Which model is correct? I am so confused.

freeztar wrote:

Visually: 1---2---3---4 (44k) 1-2-4-5-6-7-8 (88k) How does the latency change here? (assuming that latency is a measure of the time difference between input and output) What you are seemingly saying (visually): 1---2---3---4--- (44k) 1-2-3-4- (88k) Which model is correct? I am so confused.

well first off, you left out the 3 in your first visual aid, so i guess the second one would be correct by default

seriously though, your visuals are the same except you showed only the first bit of the 88.2's run in the second one? what are you trying to say here?

Your model is correct. If the latency is measured in samples, it will be half as much in terms of milliseconds at the higher sampling rate, which you can see in your second picture.

You are thinking of "latency in samples" as being some absolute amount of time. This is not true. You can have one sample per minute if you want, though of course that would be completely useless for audio. In that absurd case 128 samples of latency would be 128 minutes; try playing that in realtime. With a sampling rate of 1Hz (once per second) it would be 128 seconds. With a sampling rate of 44.1KHz (44100 times per second) it works out to 2.9 milliseconds or something like that. If you set the latency in samples, then the delay in terms of actual time will vary with the sampling rate.

In your specific setup, if you are setting the latency in ms to begin with, none of this really matters except for educational purposes. If you set it to 4 ms you will get 4 ms.

I don't know the technical details (and I don't want to know) but I know for a fact that if I choose a higher sampling rate I can achieve better latency with the same hardware.

contrast wrote:

Your model is correct. If the latency is measured in samples, it will be half as much in terms of milliseconds at the higher sampling rate, which you can see in your second picture. You are thinking of "latency in samples" as being some absolute amount of time. This is not true. You can have one sample per minute if you want, though of course that would be completely useless for audio. In that absurd case 128 samples of latency would be 128 minutes; try playing that in realtime. With a sampling rate of 1Hz (once per second) it would be 128 seconds. With a sampling rate of 44.1KHz (44100 times per second) it works out to 2.9 milliseconds or something like that. If you set the latency in samples, then the delay in terms of actual time will vary with the sampling rate. In your specific setup, if you are setting the latency in ms to begin with, none of this really matters except for educational purposes. If you set it to 4 ms you will get 4 ms.

Fisrt of all, thank you for answering my banal questions.

I *think* I understand now.
The only confusion left for me now is why latency has always been expressed in terms of ms (in my limited experience), as opposed to samples/second. I guess it's just more universal, for all sample rates?

BONES wrote:

I think that greater bit-depth [24-bit or 32-bit] is far more important than higher sample-rate but if your PC can handle both then why not work in the highest quality you can? That said, I always record/render at 44.1/32-bit.

Yep usually the same here.

Ok, here's what my limited knowledge has to say on this matter...

First of all, different converters perform differently at various sample rate. A specific ADC going up to 192kHz may sound better at say 48kHz than anything else. This is up to the hardware on what works best.

Ultra high sample rates are not meant so much to cover ultrasonic frequencies, but rather to make the higher audible frequencies sound better. So, you may not be able to hear more than 18-20kHz due to your biological limitations... but higher sampling rates may cause your upper frequency content, that you CAN hear, to sound better.

As I said, this is all hardware dependant, you may end up with a converter which sounds no different or worse at its higher sample rates.

Now, there is a real noticeable difference when it comes to processing. Simply put, digital arithmetics need as much precision as possible in order to end up with the best result. Simple as that. You wouldn't ever need 64 bit audio to satisfy your aural senses BUT... digital algorithms will benefit from such precision. Less errors and rounding in the calculations. A low precision calculation may come up with, say, 1.02 as a result. A higher precision calculation will be something like 1.0234456. Given that millions such calculations may be occuring at any given moment on a DAW, the accuracy of the numbers affects the perceived quality.

Digital synths & effects use the sample rate as a clock and a basis on which the sound is generated. Higher sample rates will make an algorithm perform more calculations in the same time frame, hence the increased cpu usage. Some algorithms, need to calculate much faster than the sample rate clock, in order to keep accuracy up. 44.1kHz may be great for your audio track, but it may not be enough for the processing algorithms to maintain high precision. This is where oversampling comes along. Oversampling makes the algorithm work at double or more the original clock (i.e. sample) rate. It works much faster and therefore has more precision in order to minimise audio artifacts.

Some synths/fx work at the project sample rate. These may benefit if you increase this sample rate and provide you with a cleaner sound (remember, greater precision). It gets more complex than that... Some algorithms (usually from older synths/fx) may not handle high sample rates well, generating a worse sound than the original one. Finally, many algorithms already employ oversampling, and automatically work at higher sample rates internally even if you are still at 44.1kHz with your ASIO driver. These will most likely sound the same if you move to say 96kHz, you will usually only get higher CPU usage and no benefit in sound.

Summing it up... high precision s/rates and bit-depths are mostly good because of the digital processing involved. Not so much because the ear MAY hear ultrasonics, or some part of your brain respond to them etc etc. It's just that arithmetics gets more room to work with... and that may or may not sound better or worse depending on what kind of algorithms you are working with.

Hope this helps... and I hope I have my facts straight.

contrast wrote:

This begs the question of what, exactly, is being lost; there are fewer samples per cycle, that means less information per cycle, so logically we are losing something. What is being lost is, and is only, the portion of the wave that contains frequencies above 22050.

Exactly. This is why the connect the dots argument (although seemingly obvious) doesn't hold water.

Personally I've never heard much difference in the ultrahigh sample rates - definitely not enough to be worth flogging my PC to death and reducing the number of FX, dynamics etc that I can use in any project.

If you have a quick enough CPU and enough RAM etc, then it may be worth the extra workload to some. To me it's not. Unless you are making extremely dynamic music that is very sparse and very delicate....I doubt most people here could actually tell any difference on a dense pop/dance/rock mix whether it was rendered highly, mixed highly or any mix of the above.

Personally I use 24bit because I can hear quite a difference during the rendering and recording stage with reverb tails - having said that, I seriously doubt if I blind tested myself with a 24bit full mix that was reduced down to the normal 16 bit, and a mix that was all done in 16bit, that I could actually hear any difference. To me the point is I like the quality at higher bit depth on soloed reverb channels when I'm mixing etc, but I don't think I actually gain any end result in the long run - my mixes all end up at 44.1/16bit for CD. So the same applies with high sampling rates. I'm already flogging my PC by doing everything at higher bitdepth...so why torture it more with higher rates. You really do have to ask yourself if you seriously hear any improvement in quality in your final CD mix.

If you're recording everything to DVD, obviously it's a different issue...but let's face it - the vast majority of all of us end up at 44.1/16bit.

The way I see it, anything above 44.1KHz is "oversampling". Oversampling is useful if effects or generators you use produce aliasing noise. So I look upon high sample rates as a workaround for sucky oscillators etc (I never use it though, I scrap the synths that produce too much aliasing noise instead...)

All this theorisation crap is so pointless. If and maybe and not sure blah.
For the 5th time, try it OUT. Vanguard is a good example. The superwave synths are good examples.
If you only have half decent monitors or headphones you will hear the difference. Whether or not you prefer the aliasing version or the alias-free version is up to everyone's taste. But unless someone's deaf above like 12khz, the difference is clearly audible.

Markus

akisd28 wrote:

DevonB wrote: You're playing connect the dots with 96,000 points or 44,100 points over a one second segment. No, you're not playing connect the dots; yes, in SOME VSTi there's a huge difference in sound between different sample rates (in my experience Lounge Lizard, Powercore 01, Superwave synths etc.; NOT in Pro53, B4 and some others), but it hasn't anything to do with the dots. Sorry, I'm not the best to explain what's happening exactly.

Ok, you tell me I'm wrong, but won't tell me why? Try again.

Sampling rate is the amount of times the audio device captures the amplitude of the incoming analog signal over a one second interval. It uses all those captured points to re-create the analog waveform. If we take 44,100 points captured over a one second interval, please explain how it's not 'connect the dots' between each sample?

Also, keep in mind higher sampling rates and high bit depths gives more opportunity to more accurately recreate a sound. Key word here is "REcreate". Even though we can represent a specific freqeuncy, it's how much closer we can get it to the original sound with higher sampling rates.

How much you'll HEAR the difference is a different story and depends on the source material, as always.

Devon

You guys seem to be a bit confused.
There's quite a huge difference between recreating material (recording and then playing back samples) and creating material (digitally syntheeizing it).

If an oscillator tries to create waveforms without bandwidth-limiting them, like for instance vanguard and the superwave synts, the result will be different depending on sample rate, regardless of which sample rates you compare. When using bandwidth-limited oscillators (limited to for instance 20KHz) the result of that oscillator will be the same for _any_ sample rate above bandwidth * 2 (40KHz in this example).

This is due to "faulty" constructed oscillators (or effects for that matter) and has nothing to do with the correctness of recording / playing back sampled material. In theory, the result when recording and playing back material in the range of 0-20KHz, the result will always be the same (exactly the same) regardless of how high above 40KHz in sample rate you go.

DevonB wrote:

Ok, you tell me I'm wrong, but won't tell me why? Try again.

You are wrong and I did explain why. Contrast then gave another explaination why you are wrong. I'll try again if you like?

DevonB wrote:

akisd28 wrote: DevonB wrote: You're playing connect the dots with 96,000 points or 44,100 points over a one second segment. No, you're not playing connect the dots; yes, in SOME VSTi there's a huge difference in sound between different sample rates (in my experience Lounge Lizard, Powercore 01, Superwave synths etc.; NOT in Pro53, B4 and some others), but it hasn't anything to do with the dots. Sorry, I'm not the best to explain what's happening exactly. Ok, you tell me I'm wrong, but won't tell me why? Try again. Devon

Both nuffink and contrast explained that better than I could.

Bunch of armchair travelling VIPs around here me thinks.

nuffink wrote:

DevonB wrote: Ok, you tell me I'm wrong, but won't tell me why? Try again. You are wrong and I did explain why. Contrast then gave another explaination why you are wrong. I'll try again if you like?

Please, do so. I'm referring specifically to how 44,100 samples taken in 1 second time interval of the amplitude of the signal is NOT playing connect the dots.

Devon

It all comes down to this: higher sample rates have much broader transition bands.

The transition band is the frequency band between the highest frequency you care about (typically 20KHz) and the actual Nyquist-Shannon frequency that is always half of the sample rate.

In the case of a 44.1K sample rate the transition band may be specified as 22.5K/20K which equals 1.125 – in log base 2 terms this value works out to 0.17 of an octave. A real-world filter that works over a transition band this small is very difficult to design. (These days, oversampling techniques are used in the DAC hardware to relax the filter requirements for any sampling rate – but for the moment lets not consider this aspect of the issue)

If we bump up the sample rate to 48K things get better. 24K/20K equals 1.2 – or 0.26 of an octave. Designing a filter to work over this transition band is not nearly as difficult. In fact, this seemingly tiny bump in sampling rate reduces the required filtering complexity by an order of approximately 2 to 1.

Even if you factor in the hardware assisted oversampling, for any given sample rate the higher the sample rate the broader the transition band will be and that means a better realization of the signal in the real world.

Jumping up to 88.2K or 96K sample rates makes filtering almost trivial. But at a cost: these rates do double the computational complexity of the entire process and some CPUs are just not up to the task just yet.

A final note: there are still people in the world who use magnetic tape to record audio. The bias frequency of the really high-end machines is approximately 400KHz – while it is not exactly equivalent, intuitively this translates into a very high sampling frequency. At 192K samples per second we are still only half of the way to this goal.

DevonB wrote:

Disagree here. Higher sample rate will give more accurate recreations, and on some VST's, yes it does make the sound a bit smoother. You're playing connect the dots with 96,000 points or 44,100 points over a one second segment. You tell me which would be more accurate and smoother? Also, 96k is going to prevent some aliasing too, at least for freqeuncies up to 48kHz instead of only 22.1kHz. Devon

When "connecting the dots" (like you do when recording digitally and then playing back) sample rates above twice the highest perceived frequency won't do you any good whatsoever.
On VSTi's (and VST's) it might matter, but that's not because of the reason you stated. It certainly has nothing to do with accuracy. When running a aliasing generator at 96KHz it will still produce aliasing, and this should be compared to running a alias free generator (which most of us prefer) which won't alias even at 500hz sampling rate.

When the content is created in the digital domain it's no longer about connecting the dots, and when we're speaking about "recreations", the best sample rate is dependant on your adc (if you yourself recorded it) and most of all, the dac. As long as you're above 40KHz that is.

DevonB wrote:

nuffink wrote: DevonB wrote: Ok, you tell me I'm wrong, but won't tell me why? Try again. You are wrong and I did explain why. Contrast then gave another explaination why you are wrong. I'll try again if you like? Please, do so. I'm referring specifically to how 44,100 samples taken in 1 second time interval of the amplitude of the signal is NOT playing connect the dots. Devon

It is connecting the dots when you convert it from the digital domain. Regardless of samplerate you can still recreate any signal below nyquist perfectly.

pj geerlings wrote:

[lots of interesting stuff]

Hm, why do you have to filter in the "transition band"? Are you speaking of ADC'ing or DAC'ing?

xRAVENx wrote:

Bunch of armchair travelling VIPs around here me thinks.

Indeed.. These are the kind of threads that remind me that this is KvR.

DevonB wrote:

Please, do so. I'm referring specifically to how 44,100 samples taken in 1 second time interval of the amplitude of the signal is NOT playing connect the dots.

Connecting the dots looks intuitively like a good idea. Connecting the dots is a good approximation when the signal you're sampling is very low frequency compared to the sampling rate. But connecting the dots isn't Correct.

Consider the case where you're sampling a full range sine wave at exactly 1/2 the Nyquist rate (= 1/4 the sample rate, = 11025 Hz for 44100Hz sample rate). If the sampling phase exactly matches up with the phase of the sine, then you'll get the following samples out:

0, +1, 0, -1, 0, +1, 0, -1...

If you "connect the dots", you'll see a triangle wave, but this isn't a triangle wave - it can't be a triangle wave, because a triangle wave at that frequency would be carrying substantial energy well above the Nyquist rate, which we know from sampling theory is impossible. Hence reconstructing the analog signal corresponding to that string of samples must be something other than "connecting the dots".

(Another example worth considering is a sine wave very close to Nyquist. If you sample that and then connect the dots to reconstruct it, what you get out looks like an amplitude-modulated triangle wave. The amplitude modulation also implies energy above Nyquist, and thus is wrong.)

The correct way to go from the sampled representation to the perfect analog reconstruction is to go through a lowpass filter with a very very steep cutoff - as close to infinite slope as you can get. Because a sharp cutoff in frequency-space implies a long impulse response in the sampling domain, it turns out that every point you've sampled winds up affecting the reconstructed curve at every point in the curve, even sampled points very far away from the portion of the curve you're considering. (Look up 'sinc reconstruction' on google if you're ready to bake your noodle.) Fortunately, the influence of each point does decrease with distance, so at some point you can stop worrying about the influence of far-away sample points.

i thought it was all nonsense before trying it.....but indeed SOME vstis do sound alot better at 96hz.
wether this is due to poor code or not i have no knowledge to judge......but the sound at 96hz is clearly fuller.
it would solve most problems if all vsti supported oversampling.
for everything else 44.1 hz seems to be just fine.

_starcraft_ wrote:

i thought it was all nonsense before trying it.....but indeed SOME vstis do sound alot better at 96hz. wether this is due to poor code or not i have no knowledge to judge......but the sound at 96hz is clearly fuller. it would solve most problems if all vsti supported oversampling. for everything else 44.1 hz seems to be just fine.

now go try some distortion fx...

stefancrs wrote:

pj geerlings wrote: [lots of interesting stuff] Hm, why do you have to filter in the "transition band"? Are you speaking of ADC'ing or DAC'ing?

Actually both

The existance of a transition band is what destroys any hope of the theoretic "perfect" reproduction (and recording).

The filter that is required (both in to and out of a digital system) must pass all frequencies of interest with no attenuation and completely attenuate all frequencies above the Nyquist-Shannon limit frequency. A transition band always exists for any sample rate because real-world filters will never have infinite cut-off rates.

The existance of a transition band seems to be the most over-looked aspect in these discussions. IMO, understanding it is very important in understanding the limits of "perfect" digital reproduction.

peace,
pj

yeah, distortion is one of those that always seems to benefit (unless you want it to sound gritty) and same goes for 99% of the EQ plugins out there.

EDIT (now corrected spelling too)

I forgot to add that a lot of old VSTi's might sound "worse" because the filter changes it's cutoff frequency and Envelope/LFO behaviour can also change when ran at 96khz. Just re-adjust the patch to sound identical to the original and all should be fine.

- bManic

_starcraft_ wrote:

i thought it was all nonsense before trying it.....but indeed SOME vstis do sound alot better at 96hz. wether this is due to poor code or not i have no knowledge to judge......but the sound at 96hz is clearly fuller. it would solve most problems if all vsti supported oversampling. for everything else 44.1 hz seems to be just fine.

Exactly. Some do, some don't. I suggest anyone curious about this bump their sample rate up to 88k or 96k and audition a few of their favorites. If you can't hear a difference then there's no point. I was surprised myself.

z3ta is also an interesting test for this. Some patches sound identical at 2x oversampling. Some sound quite different.

pj geerlings wrote:

stefancrs wrote: pj geerlings wrote: [lots of interesting stuff] Hm, why do you have to filter in the "transition band"? Are you speaking of ADC'ing or DAC'ing? Actually both [the well carried explanation here]

wtf, I knew that this trying-to-stop-smoking is seriously damaging my brain, short-term.

Borogove wrote:

DevonB wrote: Please, do so. I'm referring specifically to how 44,100 samples taken in 1 second time interval of the amplitude of the signal is NOT playing connect the dots. Connecting the dots looks intuitively like a good idea. Connecting the dots is a good approximation when the signal you're sampling is very low frequency compared to the sampling rate. But connecting the dots isn't Correct. Consider the case where you're sampling a full range sine wave at exactly 1/2 the Nyquist rate (= 1/4 the sample rate, = 11025 Hz for 44100Hz sample rate). If the sampling phase exactly matches up with the phase of the sine, then you'll get the following samples out: 0, +1, 0, -1, 0, +1, 0, -1... If you "connect the dots", you'll see a triangle wave, but this isn't a triangle wave - it can't be a triangle wave, because a triangle wave at that frequency would be carrying substantial energy well above the Nyquist rate, which we know from sampling theory is impossible. Hence reconstructing the analog signal corresponding to that string of samples must be something other than "connecting the dots". (Another example worth considering is a sine wave very close to Nyquist. If you sample that and then connect the dots to reconstruct it, what you get out looks like an amplitude-modulated triangle wave. The amplitude modulation also implies energy above Nyquist, and thus is wrong.) The correct way to go from the sampled representation to the perfect analog reconstruction is to go through a lowpass filter with a very very steep cutoff - as close to infinite slope as you can get. Because a sharp cutoff in frequency-space implies a long impulse response in the sampling domain, it turns out that every point you've sampled winds up affecting the reconstructed curve at every point in the curve, even sampled points very far away from the portion of the curve you're considering. (Look up 'sinc reconstruction' on google if you're ready to bake your noodle.) Fortunately, the influence of each point does decrease with distance, so at some point you can stop worrying about the influence of far-away sample points.

Thank you. Seems I've been grossly oversimplifing the process.

Devon

bmanic wrote:

yeah, distortion is one of those that always seems to benefit (unless you want it to sound gritty) and same goes for 99% of the EQ plugins out there. - bManic

When talking about the distortions gritty = aliasing noise.

oh, btw, "EIDT" ???

OK, how about a related question; what would have a more drastic effect on recorded [live] material moving to a higher (let's say 96k) samplerate on a prosumer card [M-Audio and the likes] or buying a high end converter [Apogee, Lucid, Lavry] and staying at 44.1k? I assume it will be the latter correct? Now can we get a bit into how the two relate to each other? I mean will I loose the benefits of a good apogee converter at 44.1k as soon as I start mixing, or will it still be there by the end (I'm guessing yes). Secondly, if all this stuff about 96k representing sine waveforms near nyquist better than 44.1k is true than why will a good 44.1k converter still outperform a mediocre 96k converter?

I'm asking because I'm looking to buy an external converter to save me from increasing samplerate as I'm hearing a common sound [a sort of graininess] in my recordings that I'm not happy with, whereas I know my mics [Rode, Blue, Shure] and pre's [Grace] are quite good. The problem of increasing samplerate is that I'm not sure how great a benefit I'd recieve from an M-Audio Delta 66 converter at 96k than at 44.1k (in theory it could sound worse), and I'd run out of CPU power quickly at mixing at time in a project with 30+ tracks with lots of compressors and EQs as well as the stock synth or two, delay, and reverb.

pj geerlings wrote:

It all comes down to this: higher sample rates have much broader transition bands. The transition band is the frequency band between the highest frequency you care about (typically 20KHz) and the actual Nyquist-Shannon frequency that is always half of the sample rate. In the case of a 44.1K sample rate the transition band may be specified as 22.5K/20K which equals 1.125 – in log base 2 terms this value works out to 0.17 of an octave. A real-world filter that works over a transition band this small is very difficult to design. (These days, oversampling techniques are used in the DAC hardware to relax the filter requirements for any sampling rate – but for the moment lets not consider this aspect of the issue) If we bump up the sample rate to 48K things get better. 24K/20K equals 1.2 – or 0.26 of an octave. Designing a filter to work over this transition band is not nearly as difficult. In fact, this seemingly tiny bump in sampling rate reduces the required filtering complexity by an order of approximately 2 to 1.

This is a great explanation! In thinking about this issue, a question came to mind: Given the importance of the transition band size, do you see any advantage to recording at 48KHz instead of 44.1KHz, then using something like Voxengo's R8brain Pro Sample Rate Converter to go to 44.1Khz?

Or, would this additional step degrade the audio and therefore offset the benefit of the higher sample rate?

_starcraft_ wrote:

i thought it was all nonsense before trying it.....but indeed SOME vstis do sound alot better at 96hz. wether this is due to poor code or not i have no knowledge to judge......but the sound at 96hz is clearly fuller. it would solve most problems if all vsti supported oversampling. for everything else 44.1 hz seems to be just fine.

It's not due to poor coding-it's due to GOOD coding.
Although it's said that the human ear only perceives up to 20k or so,I think our brains perceive frequencies way beyond that.That's what make the old Moog modulars so special sounding-they went out beyond 30k in frequency response.Put a Moog modular against your traditional hardware synth from the 80s-90s and the other synth will sound like there's a blanket in front of the speakers...even though the newre synth is going out to 20k or so...
ew

stefancrs wrote:

bmanic wrote: yeah, distortion is one of those that always seems to benefit (unless you want it to sound gritty) and same goes for 99% of the EQ plugins out there. - bManic When talking about the distortions gritty = aliasing noise. oh, btw, "EIDT" ???

Hehe, sorry.. I'm sure you knew I wanted to say EDIT:

EnzymeX wrote:

Question: Given the importance of the transition band size, do you see any advantage to recording at 48KHz instead of 44.1KHz, then using something like Voxengo's R8brain Pro Sample Rate Converter to go to 44.1Khz? Or would this additional step instead degrade the audio and therefore offset the benefit of the higher sample rate?

These days I always use 48K - but that's just me

I see Vanguard was mentioned several times in regards to sounding better at higher rates....

Oh, come ooooooooooooonnnnnnn!
The majority of Vanguard users are trying to get 3-oscillator, stacked tarnce leads at 55-voice unison mode. How the fuck can anyone hear any difference with those type of sounds even if it's recorded at 8bit/33KHz ?

pj geerlings wrote:

EnzymeX wrote: Question: Given the importance of the transition band size, do you see any advantage to recording at 48KHz instead of 44.1KHz, then using something like Voxengo's R8brain Pro Sample Rate Converter to go to 44.1Khz? Or would this additional step instead degrade the audio and therefore offset the benefit of the higher sample rate? These days I always use 48K - but that's just me

Thanks for the response, pj. If you don't mind, I have two more quick questions:

-How do you downsample/dither to 44/16?
-Do you notice degredation when you downsample (compared to, say, a straight dither from 44/24 to 44/16)?

EnzymeX wrote:

pj geerlings wrote: EnzymeX wrote: Question: Given the importance of the transition band size, do you see any advantage to recording at 48KHz instead of 44.1KHz, then using something like Voxengo's R8brain Pro Sample Rate Converter to go to 44.1Khz? Or would this additional step instead degrade the audio and therefore offset the benefit of the higher sample rate? These days I always use 48K - but that's just me Sound like a good idea for DVD compatibility also. If you don't mind, I have three more quick questions: -Do you record at 48/32 or 48/24? -How do you downsample/dither to 44/16? -Do you notice degredation when you downsample (compared to, say, a straight dither from 44/24 to 44/16)?

DVD is 96kHz/24 bit. DVCam is 48kHz though, and the reason why I'm set to 48kHz.

Devon

DevonB wrote:

DVD is 96kHz/24 bit. DVCam is 48kHz though, and the reason why I'm set to 48kHz. Devon

Thanks, DevonB

kritikon wrote:

I see Vanguard was mentioned several times in regards to sounding better at higher rates.... Oh, come ooooooooooooonnnnnnn! The majority of Vanguard users are trying to get 3-oscillator, stacked tarnce leads at 55-voice unison mode. How the fuck can anyone hear any difference with those type of sounds even if it's recorded at 8bit/33KHz ?

Sounds bloody ugly at 44khz, and does not at 96khz. Besides your genre issues, I'm sure you'll hear the difference. And even though it might be popular to bash 'tarnce' the elaborations don't exactly make your own level of expertise shine brightly.

Markus

pj geerlings wrote:

EnzymeX wrote: Question: Given the importance of the transition band size, do you see any advantage to recording at 48KHz instead of 44.1KHz, then using something like Voxengo's R8brain Pro Sample Rate Converter to go to 44.1Khz? Or would this additional step instead degrade the audio and therefore offset the benefit of the higher sample rate? These days I always use 48K - but that's just me

Very wise indeed. It'll sound better. Especially if you run a lot of tracks with lots of EQ etc, then it will sound quite a lot better.

Cheers!
bManic

YASD, yet another samplerate discussion!

For a starter, I would like to refer here:
http://www.kvraudio.com/forum/viewtopic.php?t=68114&postdays=0&postord er=asc&start=15

About Connect-the-dots
Transfer of a string of number from the digial domain to a continuous signal is not connect the dots.
Audition/Cooledit has is very good demonstration of this effect. You can zoom into the wave until you can see the sample points, plus you can see the corresponding continuous wave resulting from the sample points (be surprised!).
If you want the ultimate kick, try converting sample rates to 96 kHz or even down to 22 kHz and observe that the continuous wave hasn't changed shape (as long as no signal energy was lost).


Now the difference of cheap vs expensive sound hardware is it's ability to make the continuous wave from the sample points as good as Cooledit can do.


Why not make the same sample rate experiments with an aliasing free synthesizer, like the IBlit, PolyIBlit, Xhip or the like? Then report you can hear a difference between running IBlit at 44.1 and running it at 96?

ew wrote:

It's not due to poor coding-it's due to GOOD coding. Although it's said that the human ear only perceives up to 20k or so,I think our brains perceive frequencies way beyond that.That's what make the old Moog modulars so special sounding-they went out beyond 30k in frequency response.Put a Moog modular against your traditional hardware synth from the 80s-90s and the other synth will sound like there's a blanket in front of the speakers...even though the newre synth is going out to 20k or so... ew

i really don't believe in this "other level of listening" theory.....
I just hear what my ear allows me to hear.
I still believe its to do with code being more efficient with higher sampling rates.....but just cos it aids its calculations and helps deal with this alias stuff.we are talkin about generated sound here.i think its more to do with what the computer can hear (numbers) ....than what we can hear as humans.
if i record my voice at 44.1 or 96 it makes no difference whatsoever(no matter how good the mic is).
a dev hopefully will clear this up for us.....

I'm set to 24 bit 48 Khz with Cubase SX and an Echo Mia sound card. The difference from my old PC (same sound card) running at 16 bit and 44.1 Khz is definitely noticeable.

You simply have to dither your master down to 44.1 and 16 bit when everything is done.

Christian Schüler wrote:

Audition/Cooledit has is very good demonstration of this effect. You can zoom into the wave until you can see the sample points, plus you can see the corresponding continuous wave resulting from the sample points (be surprised!). If you want the ultimate kick, try converting sample rates to 96 kHz or even down to 22 kHz and observe that the continuous wave hasn't changed shape (as long as no signal energy was lost).

That's because what CoolEdit is showing you between the "dots" is a hermite interpolation. Sample rate conversion, when properly coded, smoothly interpolates rather than just dropping samples or repeating the values to fill in.

contrast wrote:

UvRayz wrote: Kuniklo: Could you elaborate on the logic behind that procedure? Work at 44.1k to keep the load on the computer to a minimum, bump it to 96k and render to get the best sound out of the plugs.

If your only intention is to work in a processor-friendly "draft mode", why stop at 44.1k? For draft work, it's going to be tough to even distinguish it from 22050, half again the processing. Yeah, I know, we're getting into audible frequencies now, even for 40 year olds. I prefer to lock everything in at 48kHZ, and I'm far more concerned with dynamic range than with frequency response, but that's just old-school, classical composin' me.

Why 48kHz and not 44.1? I have a master clock that works at 48kHz, and I'm very sure that sampling is in sync. And since my recording stuff can handle twice that rate, I like to delude myself into beleiving it's in a midpoint of some margin of headroom. In other words, what it can do in a flat out run, it ought to be able to do half that in its sleep. I certainly think this is true for the bottleneck where the bits have to go through the disk controller's buffer, and anything at all that has to do a bitwise operation on the sample.

48K is a nice happy medium, and happens to be an even number, which appeals to my sense of symmetry.

Now, as to whether 24 bits of dynamic range are better than 16, I definitely say yes, and politely allow anyone who disagrees to do it their way. Most people seem to be struggling to keep everything in the top 4 bits anyway.

I actually think it's best to work and render at the same sample rate. I was just explaining the thought process behind not doing so since someone asked.

I would definitely work at 24 bits if at all possible, not much reason not to unless your gear just doesn't support it.

Quote:

why stop at 44.1k? For draft work, it's going to be tough to even distinguish it from 22050, half again the processing.

22050 means a maximum frequency limit of 11025hz, which is below maximum frequency perception limit. This will make a big difference in many cases (human perception goes up to about 17-18khz). Imagine not being able to turn the cutoff knob of a filter above 11025hz, it would only allow for pretty dull sounds...

Well I haven't read the complete thread, but I hopefully can add something new and share my limited knowledge...

In most cases it's sufficient if the Synth is doing oversampling, working with n * sampling rate, internaly. The advantage is, that the calculated overtones above the audiable frequency can be filtered before downsampling to 44.1khz. If you don't filter them, even overtones in non audiable frequencies can create audiable aliasiang when inclompletely calculated.

I hope this brings a little light into things...

Summa

As if the issue wasn't convoluted enough I would like to offer this ...

For a given sample rate there is a specific number of sample points that must be used to produce a sinc interpolated 2X over-sample. This number is a function of the required flatness and stop-band attenuation. For an attenuation of -126.42dB and a response dip of no greater than 0.1 db here is some interesting data:

@44.1K 60 pts -0.09dB 2.646 MFlops
@48K 34 pts -0.08dB 1.632 MFlops
@96K 12 pts -0.014dB 1.152 MFlops

It is interesting that the number of operations for the 96K case is actually less than half that required for the 44.1K case.

Thanks. If I understood correctly, you suggest that if a synth is well crafted, there shouldn't be much advantages in having project samplerate above 44.1 kHz.

Summa wrote:

Well I haven't read the complete thread, but I hopefully can add something new and share my limited knowledge... In most cases it's sufficient if the Synth is doing oversampling, working with n * sampling rate, internaly. The advantage is, that the calculated overtones above the audiable frequency can be filtered before downsampling to 44.1khz. If you don't filter them, even overtones in non audiable frequencies can create audiable aliasiang when inclompletely calculated. I hope this brings a little light into things... Summa

AntiPro wrote:

Quote: why stop at 44.1k? For draft work, it's going to be tough to even distinguish it from 22050, half again the processing. 22050 means a maximum frequency limit of 11025hz, which is below maximum frequency perception limit. This will make a big difference in many cases (human perception goes up to about 17-18khz). Imagine not being able to turn the cutoff knob of a filter above 11025hz, it would only allow for pretty dull sounds...

I dunno. I guess it depends on what you're used to and what equipment you're using, but I'm a pretty much 'normal' dude with a normal stereo, like the majority of the population. I could easily run a low-pass filter at even 9,000 kHz and usually not notice the difference.

Sometimes I'd even CHOOSE to do so, to get rid of any offensive hissy leftover crap way up there.

Don't get me wrong, I'd never work in 22,050 anyways, just saying that there's a whole lot of frequencies up there that I could do without and not feel too much pain.

Greg

wtf? are you lot all deaf or what?

Yes.

Even my old ears noticed a difference in some cases between 44.1 and 96 - but I hardly ever noticed much of a difference between 44.1 and 48.
So, for me, working at 96 *would* make sense if (my) computers were faster.
48 would *perhaps* make sense too on rare occasions, but the hassle to convert each and every fucking sample coming from a CD or so from 44.1 to 48 is just annoying me too much for the little outcome I'd gain from this. Let alone that around 90% of my drumsamples are in 44.1, so if I was working at 48, no preview (using DR-008, Battery or the EXS) would sound correct (they all only convert the samples to the project's sample rate once you actually load them).
So I just stick with 24bit/44.1kHz.

I can, in my current host, export at 32Bit IEEE/384kHz Extra High Precision, but i only ever use 16/44.

xRAVENx wrote:

wtf? are you lot all deaf or what?

Been playing music for just almost 40 years. Hell I remember when they were still pressing 78's. Granted, it was the tail end of the era, but still...

sc_a wrote:

Thanks. If I understood correctly, you suggest that if a synth is well crafted, there shouldn't be much advantages in having project samplerate above 44.1 kHz.

You can check that with Reaktor, since it can change the engine frequency indenpendently from the cards sampling rate...

...Summa

Doubled posting, KVR behaves a bit strange today...

Quote:

wtf? are you lot all deaf or what?

No. As mentioned several times already...the end result is going to be at 44.1/16 so quite a few of us see no reason for soaking up HD space with ridiculously long files and flogging a PC to death for no audible gain at the end mix. I don't dispute that some VSTi might sound marginally better at higher rates, but it's all going to be truncated and reduced at the end.

And I have another very practical reason for not using high sample rates. Since I went up to 24bit, I find that one complete song generally takes up a complete CD for storage. Most of my tracks are around 6-7 minutes each - I also do alot of rendering to audio rather than running everything live. The last 4 or 5 projects have all consumed between 600-700MB each. If I then go to 88.2KHz I'm going to need to use 2CDs to store each project - what's the point in that, when I can't hear the difference on the final mix. Just because a facility is available, it doesn't mean you have to use it. In the same way I could get latency on my soundcard down to under 1mS probably - but why would I need to? It would cripple my system.

And personally I think aliasing is one of those things that some people go on about for no valid reason. When was the last time any of you regularly played VSTi riffs at the top of your keyboard at G7 or G8....unless you do, then aliasing is really not going to be an issue. Even with harmonic enhancers, most harmonics are still going to be within the 22KHz limit.

Not that aliasing can't be a problem...but the majority of musos into electronica just won't be playing high enough for it to bother them, and the average patches they use aren't bright enough to alias.

edit: double post.

I have a Scope Board and I find the Creamware Vectron sounds much smoother at 96KHz. Some of the other Creamware synths sound smoother and silkier and slightly less rough. However, they could sound too CLEAN ???

I think once the bog standard CPU is at 5GHz and we had tons of storage we will all be working at 96 KHz and having the same discussion about 192 KHz -- although I can't run my Scope board at 96 KHz as the DSPs get overused. I like my 44.1 KHz - it sound good enough for me.

while i can beleive their is little point to recording at 96 khz and downsampling to 44.1/16 bit their is a clear and easily heard difference between some audio recorded at 96kHz and that recorded at 44.1

i own a handfull of SACD/DVD-A's most of which are cds i already owned - they sound better - not even a hint of maybe - 100 % confident of this

Now i will concede that my stereo at 4 grand or so is not an i-pod but the difference is obvious - and its not that i can hear frequencies for dogs - its better seperation of ALL instruments, much clearer soundstaging and a better sense of space - and most importantly high frequency instruments (horns, voices) sound scarily realistic

So if you plan to record for these formats (and more and more people do) you will benefit from recording at 96 Khz - how this affects vst is dependant on how they are programmed

PS - for all the join the dots haters - explain how a 44.1 hz system differentiates between a 20 Khz sine wave and a 20 khz sqaure wave - both types have one negative and one positive point

the answer is they don't - any postfiltering when converting to analogue will make the signal sound like a sine wave whether it was one or not - at SACD 2400(!) khz sampling frequency i think you might just be able tell them apart

ericj23 wrote:

PS - for all the join the dots haters - explain how a 44.1 hz system differentiates between a 20 Khz sine wave and a 20 khz sqaure wave - both types have one negative and one positive point

Well, talking about a 22040hz or so sine vs square in theory, they will both come out as sines because what makes a square wave a square wave is that it contains frequencies above the fundamental. In fact you can construct this square wave with a 22040hz sine wave combined with an infinite (in theory obviously) number of additional sines of successively higher frequency and lesser amplitude. If you put a square wave into an additive synth (eg. chameleon and whatnot), this is exactly what it will do in fact (using however many sines it can generate).

In other words, ideally there would be no difference because there is not supposed to be, at 44.1KHz frequencies from 22050 on up cannot be accurately reproduced. You've just restated that 96KHz can reproduce higher frequencies than 44.1KHz, which I don't think anyone questions.

Quote:

at SACD 2400(!) khz sampling frequency i think you might just be able tell them apart

Assuming you can hear them in the first place and further that your speakers/headphones can reproduce them with any accuracy.

But as I said in my first post in this thread, I don't doubt that higher sampling rates, up to a point, sound better. In fact, everything I have read leads to me to believe that is the case.

And they're not 16 bit so you're not just comparing differing sample rates.

kritikon wrote:

I don't dispute that some VSTi might sound marginally better at higher rates, but it's all going to be truncated and reduced at the end.

Some VSTi sound quite a bit better at higher bitrates, and the differences fall well within the range of spectrum that will remain after downsampling. This is why some devs implement 2x oversampling within the instrument itself.

You could just avoid using any such synths but I've found that quite a few of them fall into this category.

firstly the notion that both a square wave and a sine wave at 22.1khz are both sine waves are both correct if you use the everything can be made out if sines model - alas in the real world it is totally irrelevant becuase the DAC in your audio equipment does not use such an approach - now it may be oversampled and use all sorts of filtering but the bottom line is it cannot 100% accurately reproduce the wave forms put into the upper frequency range becuase it does not have enough data to follow the output - which lets face it is a constantly eveolving stream of wave data mixed in

and lets see my cd has a linear response to 40 khz - as does my amp - and my speakers -3db is at 25 khz - with -6db up at 50 khz

i can hear it Ok thanks

my point wasn't to say that recording vsts at higher rates has particular benefits but that recording audio at higher frequencies does - and once half your project is at a higher frequency then......

but i defintely dont have the knowledge to assert that the difference once it is downsampled is audible - i'm not even that convinced that recorindg at 24 bit and then truncating to 16 for cd is particulalry audible

what i am convinced about is that your dac plays join the dots - it may check a few dots ahead to make sure it is not producing nonsense but if there arent a whole lot of dots (ie at high frequencies ) it loses information

If half your song is above 10,025 Hz, then I have serious doubts about your abilities as a composer.

ericj23 wrote:

while i can beleive their is little point to recording at 96 khz and downsampling to 44.1/16 bit their is a clear and easily heard difference between some audio recorded at 96kHz and that recorded at 44.1

No there isn't. Only if the ADC or DAC works bad at 44.1KHz.

ericj23 wrote:

PS - for all the join the dots haters - explain how a 44.1 hz system differentiates between a 20 Khz sine wave and a 20 khz sqaure wave - both types have one negative and one positive point the answer is they don't - any postfiltering when converting to analogue will make the signal sound like a sine wave whether it was one or not - at SACD 2400(!) khz sampling frequency i think you might just be able tell them apart

Again, wrong, a square wave at 20KHz will sound exactly like a sine wave to any normal human. To even hear any difference at all, you'd have to be able to hear 40KHz signals. And you don't.

The fact that SACD or DVDA sounds better is subjective - If I were the mastering engineer, I bet my orders would have been to "sweeten up" the mixes on those formats or else no one will buy them. Does that mean the mix is closer to the original? Another variable is that when someone does hear a difference at other than 44.1 - it may be that their converters are performing differently, not necesarily more accurately.

I thought not to comment on this thread further because I feel I can only loose... but I can't help.


(1)
The fact that some of the SACD/DVD-recordings sound subjectively better may be due to anything besides the sample rate. This includes bit depth, sound engineering, other equipment used, better DACs, or just flatly, different EQ used during recording...

The fact is, a 96 kHz recording does not contain any information relevant to your listening experience, that wouldn't have been reproducible with 44.1.

(2)
There is no need to distinguish between different waveforms at pitches >9 kHz. As the pitch of a tone goes higher, fewer and fewer harmonics are within audible range. Thus, the higher the pitch, the less you can tell different waveforms. All tones >9 kHz should be sines to the ear, because only the fundamental is within audible range.

(3)
Again, synths and VSTi sounding smoother at 96 kHz is due to programming artifacts. They should have sounded as smooth at 44.1 in the first place. (In an ideal world). When this is not possible, a VSTi should use internal oversampling, or give the user the choice of doing so, to avoid having to hog the system with a global 96 kHz sample rate.

Generally, this is where you can tell the cheap synths and samplers with crappy wavetable interpolation and filters from the good ones.

Also again, try alias-free synths that use the BLIT or MinBLEP algorithm, which need no oversampling at all, and should sound the same no matter what sample rate used (as long as it's >44.1)

... and happy new year to every1.

I thought I killed this horse a while ago

There are at least three measurable aspects of an audio recording that may be compromised by an inadequate signal chain. These are (1) amplitude irregularities (2) phase anomalies and (3) intermodulation distortion (and this was often completely ignored in the old days). Sadly, not much has changed for the masses …

If we restrict our discussion to steady-state, single sine wave recording and playback then the first two aspects are all that matter – in this case it is possible to provide a nearly perfect rendition of an extremely boring tune - even at 44.1K sampling rate