96k vs 44.1k... big difference

[s34nsm411]

Have any of you guys tried opening files with lots of virtual instruments in 96k before? All I have to say is WOW... synths much more open sounding and a bit less 'digital' (dont we all hate hearing that term?)

One thing that I noticed was guitars (at least through podfarm) were much different. there is a lot more going on in the upper frequencies in the 96k clip, sounds like a filter has been removed almost. I kind of feel like a small amount of body was lost though but this could be the perceived effect of hearing the extra high end

http://dl.dropbox.com/u/2093596/guitartest96.mp3
http://dl.dropbox.com/u/2093596/guitartest441.mp3

just podfarm into an impulse. strangely I had to normalize them both because the 96 version was about twice as loud... not sure why?

thoughts?

 

[Empathy]

These sound like the exact same performance. Is one an up/down sampled version of the other? Downsampling a signal will make a difference, but not quite the same way as recording at the sample you're downsizing to. Please clarify.

 

[s34nsm411]

yeah this was recorded at 44.1 and then upsampled for the 96k version

I converted an entire project actually and the biggest difference was the guitars so I isolated them for this experiment

the other stuff I had like vocals, wav sample drums and such were virtually identical

vst synths also seemed to have a change in the high end

 

[Empathy]

Yeah... upsampling is a bit of an iffy topic. I was going to comment and say that the 44.1 example sounded leagues better than the 96 example, but that's just taste.

Do you understand the basics behind sample rates and how they affect a digital waveform in general? Asking now so I don't go ahead and state a load of shit you might already know.

 

[s34nsm411]

yeah I have a pretty good knowledge of the subject. I think whats going on here though is the internal processing of the plugin is oversampled which results in a different sound, with the change in the waveform from the sampling process being pretty much negligible. guitar doesnt have much top end anyway so the conversion process probably didnt alter it too much

the same test can be done with a midi and vst synth with similar results, I need to pass out but ill post a clip tomorrow

 

[drew_drummer]

The internal processing of the plugin will indeed be oversampled - but this is possible at 44.1.khz too, if the developers implement the feature. So it isn't as clear cut as 44.1khz being inferior to 96khz.

 

[Empathy]

Consider also that as opposed to a doubling in sample rate (as you'd see from 48 to 96), you're introducing anomolies going from 44.1 to 96 due to the far more complex calculations required.

 

[Harry Hughes]

Consider also that as opposed to a doubling in sample rate (as you'd see from 48 to 96), you're introducing anomolies going from 44.1 to 96 due to the far more complex calculations required.

Wasn't that was debunked years ago?
Pretty sure I've read in many places, including this forum itself, that it has been

 

[Empathy]

Wasn't that was debunked years ago?
Pretty sure I've read in many places, including this forum itself, that it has been

Not too sure actually, haven't heard anything on the matter like that, might do some extra reading.

It does make logical sense that you might get rare 'miscalculations' when making an uneven conversion, but when you think about it, they definitely would be rare considering the processing abilities of modern computers.

 

[SimonTaddio_Qc]

I used Paint to explain why it affects the high-end the more. I hope it's clear.
The upper wave is a low frequency, and the one at the bottom is a high frequency. The vertical lines represents samples, and the red line represents the wave that has been through the AD. You can easily see why it affects high frequencies the most (i.e because it almost doesn't look at all like the original wave)
Hope that helps.
http://dl.dropbox.com/u/12811539/sample rate.bmp

 

[darthjujuu]

Wasn't that was debunked years ago?
Pretty sure I've read in many places, including this forum itself, that it has been

yea...the ol' debate of 88.2 sounding better than 96 because it's a factor of 2 conversion back down to 44.1 instead of going from 96 to 44.1 which creates logarithmically more complex math. sounds the same, debunked.

 

[BLUElightCory]

Wasn't that was debunked years ago?
Pretty sure I've read in many places, including this forum itself, that it has been

Yes, it's been debunked. Modern sample rate conversion doesn't work that way, and it's no worse going from 96 to 44.1 than from 88.2 to 44.1 with a decent converter. The "divide by 2 = better" myth is really pervasive because it intuitively SEEMS to make sense, but modern SRCs don't care. (EDIT - darthjuju beat me to it).

To the OP, what you are probably hearing (if not a placebo effect) is improved plug-in performance at 96, as the audio quality (of your raw files) can not be improved when upsampled from 44.1 to a higher sample rate (though theoretically it could change slightly, but it'd most likely be almost impossible to hear the difference).

Also, most synths use audio files recorded at 44.1kHz, which won't sound any better at 96 than they do at 44.1 (though if the synth is doing any additional FX processing you could see some improved performance at higher sample rates - but the raw audio remains "44.1 quality").

 

[John_C]

Yes, it's been debunked. Modern sample rate conversion doesn't work that way, and it's no worse going from 96 to 44.1 than from 88.2 to 44.1 with a decent converter. The "divide by 2 = better" myth is really pervasive because it intuitively SEEMS to make sense, but modern SRCs don't care. (EDIT - darthjuju beat me to it).

To the OP, what you are probably hearing (if not a placebo effect) is improved plug-in performance at 96, as the audio quality (of your raw files) can not be improved when upsampled from 44.1 to a higher sample rate (though theoretically it could change slightly, but it'd most likely be almost impossible to hear the difference).

Also, most synths use audio files recorded at 44.1kHz, which won't sound any better at 96 than they do at 44.1 (though if the synth is doing any additional FX processing you could see some improved performance at higher sample rates - but the raw audio remains "44.1 quality").

+1

but technically, synths don't use pre-recorded sounds

 

[BLUElightCory]

+1

but technically, synths don't use pre-recorded sounds

Touche! I should have said "sampler," I assumed that's what he meant at least.

 

[Revson]

I used Paint to explain why it affects the high-end the more. I hope it's clear.
The upper wave is a low frequency, and the one at the bottom is a high frequency. The vertical lines represents samples, and the red line represents the wave that has been through the AD. You can easily see why it affects high frequencies the most (i.e because it almost doesn't look at all like the original wave)
Hope that helps.
http://dl.dropbox.com/u/12811539/sample rate.bmp

That's not quite right. In the digital world there are no diagonals, only up/down/left/right so the A/D conversion would be more like a stair step around the analog signal. At low bit rates that's how you get the old school Nintendo sound and at low sample rates you lose high frequencies; it's because those stair steps are really big compared to the source signal. As you increase both bit and sample rates the stair steps become smaller and tighter around the analog signal to the point where you don't hear a difference. And that's what I learned studying electrical engineering. That and calculus. Fuck calculus.

 

[s34nsm411]

To the OP, what you are probably hearing (if not a placebo effect) is improved plug-in performance at 96, as the audio quality (of your raw files) can not be improved when upsampled from 44.1 to a higher sample rate (though theoretically it could change slightly, but it'd most likely be almost impossible to hear the difference).

Also, most synths use audio files recorded at 44.1kHz, which won't sound any better at 96 than they do at 44.1 (though if the synth is doing any additional FX processing you could see some improved performance at higher sample rates - but the raw audio remains "44.1 quality").

yeah this is exactly what I am trying to point out. originally I converted an entire project file from 44.1 to 96 and bounced it down to directly A/B with the 44.1. elements that relied mostly on already generated content at 44.1 sounded exactly the same, this includes any sampled drums/vocals/etc

what WAS most notably changed however, was the guitars and to a lesser extent the vst synths (no samplers/ROMplers were used, just synths with oscillators)

I know how fundamental it is that no quality gain can be had by simply upsampling a wav file. That is why this test is such a mind blower to me, that given the same input signals these vsts can sound much different by simply changing the sampling rate at which they process the audio. And again the difference in almost everything was negligible EXCEPT for any amp sims and vst synths

I used Paint to explain why it affects the high-end the more. I hope it's clear.
The upper wave is a low frequency, and the one at the bottom is a high frequency. The vertical lines represents samples, and the red line represents the wave that has been through the AD. You can easily see why it affects high frequencies the most (i.e because it almost doesn't look at all like the original wave)
Hope that helps.
http://dl.dropbox.com/u/12811539/sample rate.bmp

im not sure what you are trying to point out here? it is true that higher frequency content will be harder to reproduce accurately, but both files in my test are good ol 16 bit 44.1 mp3s yet sound very different. the only change is the sample rate at which the amp sim plugin was operating when I bounced. It is not possible that you are hearing any magic mojo above 20,000hz. Even if my test files were 44.1 vs 96 wav, I highly doubt anyone would be able to tell unless they were a bat (other than the clear difference in the content below 20khz)

 

[John_C]

That's not quite right. In the digital world there are no diagonals, only up/down/left/right so the A/D conversion would be more like a stair step around the analog signal. At low bit rates that's how you get the old school Nintendo sound and at low sample rates you lose high frequencies; it's because those stair steps are really big compared to the source signal. As you increase both bit and sample rates the stair steps become smaller and tighter around the analog signal to the point where you don't hear a difference. And that's what I learned studying electrical engineering. That and calculus. Fuck calculus.

true

at least at the point of DAC it's not a join the dots operation. This is the correct interpretation, the original signal is in grey and the red is the signal after ADC and then DAC. (before any low pass filtering, which of course all audio DACs will use)

edit: All this topic is demonstrating is the advantage of oversampling, this is a common option in many plugins with pretty obvious results. It's not controversial or surprising.

 

[s34nsm411]

true

at least at the point of DAC it's not a join the dots operation. This is the correct interpretation, the original signal is in grey and the red is the signal after ADC and then DAC. (before any low pass filtering, which of course all audio DACs will use

edit: All this topic is demonstrating is the advantage of oversampling, this is a common option in many plugins with pretty obvious results. It's not controversial or surprising.

actually the red line is what it looks like after ADC. the DAC process turns it back into a smooth waveform. I forgot what the name of it is exactly but a single smooth wave is placed at each sampling point and when they are all summed together it results in pretty much the exact same waveform being reproduced (not the same as lowpass). This is of course, unless your original waveform has higher frequency content than 2x your sampling rate

 

[John_C]

actually the red line is what it looks like after ADC. the DAC process turns it back into a smooth waveform. I forgot what the name of it is exactly but a single smooth wave is placed at each sampling point and when they are all summed together it results in pretty much the exact same waveform being reproduced (not the same as lowpass). This is of course, unless your original waveform has higher frequency content than 2x your sampling rate

sounds like i need to do some more research........

 

[s34nsm411]

sounds like i need to do some more research........

nevermind you were right it IS a lowpass

http://en.wikipedia.org/wiki/Reconstruction_filter

but in the end it gets smoothed out anyway

For some reason I remember learning it in EE class as the sampled source being a series of impulses (dirac delta functions) and you then convolve that with a smooth analog wave that is two samples wide and presto you have your resulting smooth analog signal. Any EE guys know what im thinking of?