Don't let Live resample your audio!

[teknobryan]

so pratically, what does this mean??

Should I route the audio out of my audio interface (FF800) and back in?

[wilxon]

I thought the only reason 44.1 was ever used was back in the day when memory was very expensive, when CD's first came.

The decided to look at the sampling extinction frequency which is half of the sampling rate - IE sample rate of 100 HZ = silence at 50 Hz.


As human hearing has a tendency to stop at 22KHz, they thought they would double it so that it took the extinction frequency out of human range and add a little on which was the .1

hence 44.1

Not ideal, but the very minimum of specification needed.

[wilxon]

so pratically, what does this mean??

Should I route the audio out of my audio interface (FF800) and back in?

If you want to change the sample rate of your audio file - Do it in Audition.

[wilxon]

Really though - if you record a file using a low sample rate,

then changing the sample rate up will not change the sound of the file - so no added benefit unless you are changing the sample rate to slot that file into a project using the higher sample rate.


Better to record at higher sample rate and then resample down later.

[Spindrift]

Just a few comments relating to the points brought forth here:

Sure the main thing is how it sounds, but I guess no one wants SRC to give a certain colouration to the sound.
It just should change samplerate but keep the original as intact as possible.
If you want to colour you mix use a plug designed to do what you want.

Looking at graphs is not particularly helpful for finding out how a product sounds, but in this case it saves you the hassle of getting hold of a lot of different software, doing a lot of ABX testing to find out which one sounds best.
Looking at the graphs should give you a good idea of which products have the most transparent SRC and which one should be safe to put most material through without nasty surprises.

A lot of plugs sounds better when working at a higher samplerate, but if it's a quality product it should have internal oversampling and then it should not really be an issue.
But if you have the spare power to go 88.2 or 96k when mixing down it might give you a small improvement in some cases even if your recorded audio is 44.1k.
Upsampling is not at all as problematic as downsampling...the issues one can see on infinatewave's graphs has to do with aliasing issues and if there is no frequencies to be removed, as the case when upsamling, then there is no aliasing.

As to 88.2 being better than 96k that seems to be a myth. Sure one would think it makes sense, but in fact the actual SRC processing happens at something like 14Mhz so it's not the case that you deal with simple integers because you use 88.2k

In the end I wouldn't worry about using Live's resampling too much. If it interrupts my work flow too much to use an external app any small gain in sound quality might not be worth the hassle.
But like they say "many rivers small..."
If you want to be serious about engineering you should at least have an idea about as what factors can impair your sound, even if it's only by a very small amount.
If you in all steps in the chain think that you can get away with any solution that is not blatantly obvious decremental to your ears your end result will be suffering.

[Angstrom]

As to 88.2 being better than 96k that seems to be a myth. Sure one would think it makes sense, but in fact the actual SRC processing happens at something like 14Mhz so it's not the case that you deal with simple integers because you use 88.2k

I'm not sure that I agree on this, do you have a reference for me to read?

Because here's how I understand it.
For me the issue is not with the speed of the calculation, but the decisions the calculation has to make.

An example: Lets imagine a wave which at 44.1 plots as
[-0.5, -0.2, -0.3, +0.5, +0.25 ]

those 5 values have to be interpolated between in the upsample function.
So if we choose a linear interpolation to make things easy, 44.1 -> 88.2 means expanding the table and inserting values 88.2/44.1 = 2, so
for 88.2 we need to double the table. We can do this reliably by inserting values between our existing samples.

Here's that same table, upsampled from 44.1 to 88.2(interpolated values in italic )
[-0.5, -.35, -0.2, -0.25, -0.3, 0.2, +0.5, +0.375, +0.25]

now that was an easy calculation and although the linear interpolation will sound a little rough, it is reasonably accurate because we kept our original sample numbers and simply added ones between those points


But now we try to do the same for 96khz 96/44.1 = 2.17687075
The new table needs to be 2.17687075 times the length it originally was, it was 5 samples long ... it needs to be 10.8843537 samples long when we are done.

Of course, computers can add up OK, they can do the hard sums as quick as you like - but the problem here is - we will have to discard our original five samples because they will not fall on the sample points of 96khz. The output we get will be generated solely on the interpolation curve which we chose and the quality of the software. Which, as we can see - is often not all it should be!

so I am not so sure that 44.1->96 or the other way around is quite as foolproof as 44.1->88.2

[marky]

As an aside I'm wondering what the quality of SRC is for sample players like Kontakt, Battery, etc - clearly the same issues might exist and could easily affect an entire mix if so much of the source material is being up or downsampled...

I'm certainly going to pay more attention to the sample rates of my sources in Ableton, though for the most part I stay at 44.1k, for some of the reasons that Angstrom pointed out earlier, plus wasted CPU cycles to boot.

I don't really consider Ableton a top-flight DAW so this doesn't really shock me.. but it's a bit disconcerting that Logic Pro suffers from the same issues.

[evernaut]

Because higher sample rates produce better results when processing. Also, you can't hand in 44.1khz to be mastered - not in a professional environment anyway. And working in real-time at anything higher than 44.1 is very hardware intensive.
So for best results, you work at 44.1, render at a highest rate possible (actually 96k is probably high enough, but there's no reason not to go higher)

Upsampling from 44.1 to 96 will not make any difference. You can't add anything that isn't already there.

And I've handed in plenty of projects for professional mastering that were recorded at 44.1/ 24 bit.

[Tarekith]

If you render at 96k you can actually push the aliasing out further from the range of human hearing (ie, the strongest aliasing is now closer to 48k instead of 22k), and then when mastering and subsequently downsampling, a lot of that will get filtered out by the anti-aliasing filters. It also makes a difference if you're rendering softsynths too, for this same reason IMO.

You're right, you can't add anything, but you can push the aliasing further out of the range of human hearing. IMO, this is where the benefits of higher sample rates come into play, much more so than being able to capture higher frequencies, etc.

[Angstrom]

but you will still get interpolation issues on any samples involved

[Tarekith]

I guess the question then is what do you find more intrusive, aliasing or interpolation noise?

We can never win, can we?

[Nod]

Some of you golden eared boffins might also find this freebie useful....Win, OSX and Linux:

http://www.lcscanada.com/audiomove/

Very high end offline resampler - some say better than R8Brain.

[Tone Deft]

As human hearing has a tendency to stop at 22KHz, they thought they would double it so that it took the extinction frequency out of human range and add a little on which was the .1

clarification, it's that NOBODY can hear above 22kHz, hence the standard. in reality it's more like your hearing stops at 16-19kHz. I'm 34 and haven't abused my ears that much and mine stops at 17.2kHz. you guys are throwing numbers around but it's your ears that matter. spend some time listening to artifacts and tone generators to understand what's behind these numbers.

people are completely overlooking the data. anything in purple is at -100dB or lower, anything in white is -40dB or higher.

the artifacts are there but don't really matter.

it's one thing to be aware of the numbers, it's another to know what the ramifications are. do you REALLY know what a 15kHz tone sounds like? do you know the range of your own voice? do you know how quiet -40dB is? talking numbers is one thing, knowing what's behind them is another.

I seriously doubt most of us are good enough at mastering that any of this REALLY matters.

but it's always good to go around and kick the tires.

[evernaut]

If you render at 96k you can actually push the aliasing out further from the range of human hearing (ie, the strongest aliasing is now closer to 48k instead of 22k).

What aliasing? If you track at (and then render at) 44.1 kHz/ 24 bit there won't be any aliasing issues. All you have to do is add dither when you drop the bit depth for cd.

I honestly can't see any benefit from resampling a 44.1 kHz file to a higher rate post-mixdown. Downsampling is a very different issue and this thread highlights Live's real problems there. But when you can get free software like R8brain, it's no biggie really.

[Tone Deft]

I guess the question then is what do you find more intrusive, aliasing or interpolation noise?

We can never win, can we?

aliasing and interpolation are two different things. interpolation is a mathematical algorithm, aliasing is a result of the math.

sorry to be an ass, just talking shop.