ok I m not sure anymore : sample rate

[UKRuss]

Interesting. Bit depth I am au fait with.

Tell me more about sample rate. I have read some stuff so I understand the exact multiples issue but what real benefit is there in recording live instruments at 88.2 for example? lets forget about the 'use your ears', for the moment. Whats the science?

Also, if I am understanding correctly then, if the project is at 88.2 and 24bit and you wanted to render for CD you would render at 44.1 16bit with dither.

But, don't you simply lose all the 88.2 goodness at that point? So, one needs to resample all the stuff running through plugs at 88.2 to a new audio track before final render and THEN export at the lower sample rate, right?

edit. Also, plugs that have oversample options are already swinging themselves up beyond 44.1? Whats happening here?

[davepermen]

if you record on 88.2 and then render to 44.1 you lose exactly half of the information, indeed (if bitness doesn't change). but if NOT, if you only do slight warping, have some effect that can use the additional signal detail for more precise computation of the effect, etc...

but yes, if you record and one-to-one flatten, you gain nothing. in every other case, you might. the things you lose are storage space, and performance.

edit: guess: those plugs interpolate the other 44.1 steps between the sampled ones to generate the 88.2 sample stream. and then feed that into the effects (with warping being just another effect). now problem is, it does create a more detailed signal stream, but does not contain more actual signal INFORMATION. that, you only get by actually recording at 88.2. in the other case, it does something that you might know from other businesses: upscaling. like those dvd-upscalers for hd-tvs.

[leedsquietman]

With a good SRC you don't lose much information (certainly not half) because once you get past 48 Khz, there isn't much being captured which is audible anyway. This is one reason why people argue so much about the merits of higher sampling rates. You are basically losing things up above 22 Khz and many humans hearing is shot and can't hear beyond 17 Khz. Even the golden eared with perfect hearing struggle to hear above 20 Khz.

However, the fact that some plugins oversample/upsample and aliasing can take place because they have to be internally sample rate converted and sometimes dithered (if the project is being rendered in 24 or 16 bit instead of 32 bit float), this can sometimes be perceived as a subtle difference, even if mostly on some frequency chart printout rather than audibly.

Try it for yourself is really the only way to decide if higher sampling rates are best.

Also get yourself a good SRC, DAW Sample Rate Conversion is not usually out of the top drawer.

Ozone's SRC and Voxengo's R8 Brain are really good for this.

BTW Russ (and others) I recommend you read Bob Katz's ''mastering bible' 'Mastering Audio : The Art And The Science (2nd edition)' from Focal Press, this gives you all the information you need to know and much more beyond. It's not an easy read but a man of your experience should get it and it's amazingly in depth over this and all mastering issues.

[alex.the.forge]

yeah, like leeds said re: nyquist vs hearing - let's put it another way - if you are recording for people under 20 then it may be worth it, but most of us crusty old buggers over 30 have lost a good 5kHz or so from ageing or abuse. I'm much more interested in whether I can feel it from the sub and whether it has a good vibe. All that other shite is for the birds and my dog who hates my music anyway.

[ChrisIhao]

@UKRuss: Ok. I'm going out on a limb here, but I think the greatest advantage is that you can capture some spectre of the audio with a greater clarity. For instance, when using a convolution reverb I hear a great difference between 44/48 and 88/96. The tail of the reverb is clearly much more defined and "crisp" with a higher samplerate. Those who say you cant must have other ears than me, because its just so audible imo. It COULD be my digital converters on my audio card is somewhat off, but I highly doubt that.

Anyways, if you record the sound in 88/96 and then downsample, it still retains some of the crispness and detail. Dont ask me why, it just does in my experience. In particular this is true, like others have said, when dealing with one or more complex digital effects. Obviously many of these frequency areas or not supposed to be audible, but perhaps these frequences somehow affect the audio image as a whole? Its like taking a digital picture. You include more details when using higher resolutions, and if you downscale them they are still there, although somewhat blurred or whatever.

Now, take my comment with a grain of salt. I'm really not that deep into audio theory. However, having discussed this to death on other daws' forums these are my personal conclusions so far. Some say that the greatest advantage to higher sample rates is a better listening experience while working (playback hz, not record hz), and I think they are correct. For bedroom producers, like most of us are, the advantages may not outweigh the "costs" in terms of disk space and cpu load.

For me its just not worth going for higher hz. I run at 44/24 and can pretty much do away with everything at 5 ms latency. At 96/24 my system starts choking in most daws I've tried whenvever I introduce a few complex effects on top of 7-8 channels. Instead of bungy jumping between 128, 256 and 512 buffer settings I'll rather use a setup which works in all instances, and require as little fiddling around as possible (for me, thats creativity killer #1).

[ikeaboy]

yeah, like leeds said re: nyquist vs hearing - let's put it another way - if you are recording for people under 20 then it may be worth it, but most of us crusty old buggers over 30 have lost a good 5kHz or so from ageing or abuse. I'm much more interested in whether I can feel it from the sub and whether it has a good vibe. All that other shite is for the birds and my dog who hates my music anyway.

Exactly how I feel. High hats and cymbals are great and all that but unless I record a real kit 44.1 is what I output so it's what I'll work with

[dum]

Interesting. Bit depth I am au fait with.

Tell me more about sample rate. I have read some stuff so I understand the exact multiples issue but what real benefit is there in recording live instruments at 88.2 for example? lets forget about the 'use your ears', for the moment. Whats the science?

The benefit in recording at 88.2 might not be immediately apparent by simply comparing it dry with an identical recording at 44.1, but it will become very apparent once you start processing the recording in the digital realm - at the same sample rate of the recording, in otherwords recording at 88.2 and processing at 44.1 isn't going to do you much good. Certainly in my experience, using 3rd party vsts.

Aside from recordings, quite a few of my vst instruments benefit vastly from increased sample rates - and very obviously so (no 'golden ears' required)
I will also often work at sub 44.1 rates, for the differences in timbre and tone

Also, if I am understanding correctly then, if the project is at 88.2 and 24bit and you wanted to render for CD you would render at 44.1 16bit with dither.

But, don't you simply lose all the 88.2 goodness at that point? So, one needs to resample all the stuff running through plugs at 88.2 to a new audio track before final render and THEN export at the lower sample rate, right?

No, you don't lose the goodness unless your sample rate conversion is shit. I use izotopes SRC and that's the bees-knees, I use it in WAVE-EDITOR by AudioFileEngineering. You own Ozone, Right ? ... I'm not certain but I believe you should be able to use that for SRC, using izotopes much heralded SRC algos.

If you have any experience in graphic design...all of this would make perfect sense to you. Infact, I have seen graphics analogies of the sample rate issue in the past... which helps explain it. But really, using your ears will convince you more than such analogies... your ears will tell you if there is an improvement worth the CPU overhead, or not.

So, just to be clear, I'm not trying to foist my preferred working methods on anyone here in the thread.

[davepermen]

With a good SRC you don't lose much information (certainly not half) because once you get past 48 Khz, there isn't much being captured which is audible anyway.

which is why i was talking in my original sample about stretching some sample to 4x the time. there, you stretch the stuff that is inaudible into audible range, and thus in information that changes the signal.

and yes, you lose exactly half the information. even if, by just listening to it, you can't process that other half of information, doesn't mean it's not there (now i know, some call it 'data', then, and not 'information'. as a programmer, everything is information, because that only gets determined at the end user, out of the programmers control)

[dum]

a good sample-rate-converter will know what half of the information to bin

that's the point of choosing a good SRC

[Tone Deft]

and yes, you lose exactly half the information. even if, by just listening to it, you can't process that other half of information, doesn't mean it's not there (now i know, some call it 'data', then, and not 'information'. as a programmer, everything is information, because that only gets determined at the end user, out of the programmers control)

when a DSP algorithm is created the author can make the decision to band limit the process to save processing power, cut out noise, whatever. BUT the decision has to be made to leave THE ENTIRE audio path to not be band limited. you keep talking about more info being in the signal, but all that extra info is above what we can hear. I don't buy into this.

nobody's commenting on the audio files. those errors are egregious.

it'd be nice to hear from an Abe.

a good sample-rate-converter will know what half of the information to bin

that's the point of choosing a good SRC

funny to read you trying to join the conversation.

it doesn't know which half it just does it right to begin with. there is no 'half.'

[dum]

dave is the one going on about 'halfs' as if it's meaningful, I only stuck to his choice of words. My point is bang on.


I don't need to 'try' and join the conversation by the way. If I have said something you think is factually wrong, go ahead and confront me on it. Experience will prevail.

[Tone Deft]

dave is the one going on about 'halfs' as if it's meaningful, I only stuck to his choice of words. My point is bang on.


I don't need to 'try' and join the conversation by the way. If I have said something you think is factually wrong, go ahead and confront me on it. Experience will prevail.

hopefully common sense will prevail.

anyone fancy taking a crack at the experiment this guy tried to do?

[UKRuss]

Hold on boys, hold on, just one more minute then you can go at it.

Dum, so as an example of what you do, you may use the sample rate creatively as an effect on certain VST instruments but also becasue some effects and instruments simply hit their sweet spot at higher sample rates.

Then you render a high bit depth high sample rate stereo file, dump it into wave editor and run it through Ozones SRC to end up with a high bit depth, 44.1 stereo file ready for the mastering?

Leeds, Alex, Chris, Dave, thanks for the input. Bob Katz here I come...again. I think thats where I got my now clouded initial memory about the exact multiples thing.

[dum]

a good sample-rate-converter will know what half of the information to bin

that's the point of choosing a good SRC

it doesn't know which half it just does it right to begin with. there is no 'half.'

btw, can you elaborate on this. Besides the obvious 'half' misnomer, dave was talking purely in terms of data there.

what do you mean 'it just does it right to begin with' ?
Converting the sample rate of an 88.2hz wave file to 44.1hz wave file does indeed bin information. And the quality of the algo will determine what information is binned.

So...what exactly are you talking about here ?

[UKRuss]

DAMN!