Which Sampling Rate - 44.1 or 96?

[masteringhouse]

Found this post by Bob on Gearslutz:

http://www.gearslutz.com/board/742730-post4.html

 

[SouthSIDE Glen]

Well, I give up. I can't find that paper now to save my life. But it definitely went into deeper explanation of what Lavry was only referring two in those two links; the concept that the higher the sample rate is, the less accurate the actual sample will be, or something like that anyway.

Oh, well, I won't push it because I can't find that article. But I will say that the day I believe that Katz can hear and ID the difference in aliasing products between the different sample rates alone the way he claims in that last post is the day that I believe that Haley Joel Osmet really CAN see dead people.

G.

 

[masteringhouse]

Oh, well, I won't push it because I can't find that article. But I will say that the day I believe that Katz can hear and ID the difference in aliasing products between the different sample rates alone the way he claims in that last post is the day that I believe that Haley Joel Osmet really CAN see dead people.

G.

I can't say that I can hear all of the differences that some MEs claim, in many cases it may just be a case of puffing their chests out. This is actually one where I can hear a difference though if enough processing is done. Just a plug here or there is not very noticeable, but when added up between 24 tracks and several plugs, then A/Bing with the same tracks at a lower rate versus higher, yep. It can be subtle though, maybe about the same difference as truncation versus dithering (though a different sound).

As mentioned previously, there are more profound factors in creating great sound and a great mix. Just to get things to "tape" a series of compromises often need to be made. Choice of mic, preamp, room/working environment, recording meduim, mic position, etc. are all a compromise to some degree on the original sound and it's a lot to think about at once as well as anticipate the outcome. For mastering though where the recording and mixing process is complete (and "damage done") compromises are usually less tolerated.

Again, if your DAW and converters can handle it you're better off. If not, stay at the rate that works best.

 

[giraffe]

More than one high-end converter designer has said that if you hear a difference between 44.1k and 96k, your converters are broken.

not exactly "hi end converters" but i just (like yesterday) got a echo audiofire 2 in the mail, for some location stuff.
most reviews of their conversion are good.
any way.


the setup of the software control panel makes it easy to swap between sample rates in real time.

i was sitting and talking into an sm57-RNP
and swapping between 44.1 and 96k the difference was huge.
no, i'm not trying to be cool here, i'm sure anyone could have heard it.

difference mostly in the low end (to my ears) sound at 96 was much more "focused" and could has possibly be described as a little thinner.
but the difference was as plain as day.

it's brand new, and seems to work as advertised. i don't think any thing is wrong with it, or at least i hope not

maybe i'll record the test (the same way) at both sample rates and see if NL5 will host the files (in a aiff of something)

am busy, but will try to get it done as soon as possible.



p.s. is NL5 still around?

 

[NL5]

p.s. is NL5 still around?

I think he is.........

 

[Farview]

not exactly "hi end converters" but i just (like yesterday) got a echo audiofire 2 in the mail, for some location stuff.
most reviews of their conversion are good.
any way.


the setup of the software control panel makes it easy to swap between sample rates in real time.

i was sitting and talking into an sm57-RNP
and swapping between 44.1 and 96k the difference was huge.
no, i'm not trying to be cool here, i'm sure anyone could have heard it.

difference mostly in the low end (to my ears) sound at 96 was much more "focused" and could has possibly be described as a little thinner.
but the difference was as plain as day.

it's brand new, and seems to work as advertised. i don't think any thing is wrong with it, or at least i hope not

maybe i'll record the test (the same way) at both sample rates and see if NL5 will host the files (in a aiff of something)

am busy, but will try to get it done as soon as possible.



p.s. is NL5 still around?

That would be the difference of your converter at the two sample rates. It probably isn't the sample rate that makes the difference (especially in the low end), it's what your converter sounds like at that rate... there's a difference.

It could also be a marketing gimmick, kind of like how a BBE in bypass is 'duller' sounding than if you just take it out of the chain...

 

[TimN]

I record at 24/48--is that bad? And, good God, how has it come to pass that the MP3 is the method by which so many music consumers listen to their music? I'm thinking that the relative cheapness of megabites will cause a new demand for higher quality formats, but it seems like the MP3 and its crappy quality are here to stay. Of course, I thought the cassette would be around a lot longer than it was.

 

[CoolCat]

I get confused on, not only MP3, but all these "media players" too...are they all the same because they sure don't sound the same,imo?

24/48...then to MP3, then to some crap media player...then to

 

[XLR]

I get confused on, not only MP3, but all these "media players" too...are they all the same because they sure don't sound the same,imo?

24/48...then to MP3, then to some crap media player...then to

That's the funny part of audio production. We work so hard to capture and process our music with the highest fidelity... and 95% of listeners use earbuds and iPods. Or desktop computer speakers.

Truthfully though, higher fidelity in production certainly makes for better sound in whatever the end listening environment.

 

[CoolCat]

but like the TimN question...why isn't the industry pushing for 24/48 then?

just think they could re-sell everything all over again at 24/48!! Isn't that how its supposed to work? and we get better audio?

like Victrola's,to reel to reels, to mono to stereo albums, to 8-crap, to cassette, to digital cassette to minidisc, the CD, the MP3 players...to 24/48 players?

or will it get worse! maybe the industry will go to an even lower quality sound than MP3!

 

[SouthSIDE Glen]

like Victrola's,to reel to reels, to mono to stereo albums, to 8-crap, to cassette, to digital cassette to minidisc, the CD, the MP3 players...to 24/48 players?

or will it get worse! maybe the industry will go to an even lower quality sound than MP3!

They're already trying a successor to CD and to all PCM schemes altogether with the SACD and DSD formats. Efforts are also being made to find an agreed upon losless compression format to replace MP3.

SACD/DSD just has not caught on yet, and it could go either way as to whetehr it ever will or whether the Next Big Thing (whatever that may be) will hopscotch it. A big part of the problem is that the vast majority of the unwashed public - read: the consumers who buy the stuff - think that 44.1/16 CDs and even standard MP3s sound just fine. The switch over to SACD is apparently just not worth the format change for most people.

A replacement for MP3 may face a slightly different obstacle; lossless compression will most likely mean noticably larger file sizes than 180-200k MP3. Sure, storage space is not the issue it was 5 or 10 years ago, but when Joe Punchclock sees marketing bullet points saying that his mePod can hold 1000 MP3s but only 650 new format files, unless the new format just absolutely blows away the sound of MP3 - which in his ear buds or mePod docking station it's probably only marginally going to do - he's more likely to want to just stick with his MP3s and get the higher storage rate.

That MP3 thing is very similar to the old VHS vs Beta wars. Amongst engineers and videophiles, Beta was almost unanimously considered the superior format as far as picture quality and mechanical design. But the public overwhelmingly preferred VHS. There were a couple of reasons for this, not the least of which was the sheer number of different brands (many of them very American sounding) that Matshusita VHS machines where sold under, which gave VHS a huge shelf space advantage at the dealers. But the #1 BIG reason the custys went VHS was simply this: recording and playing time. They could stick a 6 hour (and later 8 hour) tape in their machine, which was signifigantly more than Beta offered. The fact that most VHS machines looked like absolute shit in 4- or 6-hour mode compared to your average Beta just didn't bother rmost people; they wanted more hours per tape, and were willing to put up with fuzzier pictures to do so.

G.

 

[giraffe]

user error.....

when i first tried this test i had cubase running in the background (not doing squat, just open) and when i tried it again today the results were not the same
(unless i opened cubase again)
in that, when running the audiofire control panel and switching between sample rates i still *think* i could hear a bit of a difference, but nothing obvious like when i had another program in the background effing things up.

i went through with my rudimentary test any way, and the differences between the 2 takes (even if i was very careful) were too big to hear the difference between the sample rates.
the only time i thought i could hear a difference was when i was switching in real time between sample rates in the control panel (with no other programs open this time) holding a note.
(there is a light click followed by the change in sample rate)
the difference was not big, and could be in my brain.

p.s. i did this with a decent cold. just fyi

-me

 

[LeeRosario]

This one is either for Tom, Massive, Fairview or for the experienced engineer in general...

Just throwing it out there for discussion:

If someone wanted to hear the difference, beyond a reason of a doubt, couldn't you just take 2 copys of the same waveform, one at the target sample rate and one at a downsampled rate, align those and flip your phase switch?

Whatever dosn't cancel out reveals whatever is missing in your signal? Perhaps whatever isn't audible would reveal itself on a spectragram or spectrum analyzer of some sort?

Also, what do you think about choosing sample rates for a specific target medium like say someone reaching for 44.1 with the intention of releasing material mainly on CD or MPEG versus 48 or higher for more digital and "high def" applications like DVD audio, HD DVD or Blue Ray?

 

[crosstudio]

LeeRosario: not a bad idea. if you have no signal difference you have a reasonable but not conclusive proof. If you have differences they could be attributed to the software used to downsample.

but i don't think that tells the full story. Here is what I think puts this argument to bed.

Record a complete song with X tracks all recorded at 24/88.2 and mix down to mono with all the necessary plug-ins etc.

Now, take each individual track and downsample it to 44.1 and mix down to mono with the exact same volumes/gains/plug-ins etc.

NOW IF YOU CAN ALIGN THE MIXES flip the phase on the 44.1 mono mix against the 88.2 mono mix and see what the difference is.

Because the $100,000 question isn't whether one track makes a difference, but what is the cumulative effect of having the recording and all of the plug-ins processing 88.2 verses 44.1.

 

[danny.guitar]

This one is either for Tom, Massive, Fairview or for the experienced engineer in general...

Just throwing it out there for discussion:

If someone wanted to hear the difference, beyond a reason of a doubt, couldn't you just take 2 copys of the same waveform, one at the target sample rate and one at a downsampled rate, align those and flip your phase switch?

Whatever dosn't cancel out reveals whatever is missing in your signal? Perhaps whatever isn't audible would reveal itself on a spectragram or spectrum analyzer of some sort?

I don't fall into the Tom, Massive, Fairview, or experienced engineer categories but I have to say that's a pretty clever idea...

Anyone not too lazy to try it?

 

[SouthSIDE Glen]

If someone wanted to hear the difference, beyond a reason of a doubt, couldn't you just take 2 copys of the same waveform, one at the target sample rate and one at a downsampled rate, align those and flip your phase switch?

I do not claim to be in the same class as those boys, but here's how I see it...

First, that test as described - if I understand it correctly - would IME include any artifacting added by the SRC during downsampling, and therefore would not show the pure difference between the sample rates. Second, it would not actually be measuring differences in the sampling, since there is only one sample rate used - i.e. one sample taken - at the initial conversion.

It seems to me the closest way to run the test would be to actually split the analong input signal into two converters of the same make/model, each one set for a different sample rate, so that one is actually *sampling* at two different frequencies. Then, just to try and cover possible differences between the two manufactured boxes, repeat that test, but flip the sample rates on the two boxes. that is, if the first test had converter A at 44.1 and converter B at 96, the second test would have converter A at 96 and B at 44.1.

Then you'd at least be able to compare the two resulting digital tracks without adding any extra artifacts via sample rate confersion (up or down), and use test B as a partial control to see if the results may be skewed by a difference in manufactue quality between the two boxes.

G.

 

[masteringhouse]

Because the $100,000 question isn't whether one track makes a difference, but what is the cumulative effect of having the recording and all of the plug-ins processing 88.2 verses 44.1.

Exactly!

Remember that internally the sample rate of a converter is usually much higher than the final output. It's not until it goes through the decimation process that it is converted to it's final sample rate. The difference during A/D conversion between various sample rates should be less noticeable that cascading non-linear processing with the same sample rate as in the above.

LeeRosario: not a bad idea. if you have no signal difference you have a reasonable but not conclusive proof. If you have differences they could be attributed to the software used to downsample.

Also true. I'm not sure how you can get around this issue entirely. Before experimenting as suggested, we have to be sure we are measuring the correct "thing".

Here is a very good thread on up sampling:

http://recforums.prosoundweb.com/index.php/m/167396/4063/?srch=upsample#msg_167396

G. There are some posts by Dan Lavry that may help to clear up some of the issues you were bringing up. It's confusing I know. The points that you brought up seem to be contradicting what he is saying here. But again, the difference is are we talking A/D D/A conversion or basic DSP processing.

 

[SouthSIDE Glen]

Also true. I'm not sure how you can get around this issue entirely. Before experimenting as suggested, we have to be sure we are measuring the correct "thing".

This is why I suggested sampling the identical source twice. This way one insures all they are comparing are the actual sampling process.

Even then there is one major shortcoming in such a test (beyond the SRC pollution):. Such tests will only test that particular make/model of converter. Some converters seem to "favor", or work best, at one particular sample rate or maybe have one sample rate that doesn't sound as good as the other two, because of the converter's design. Such a test can only measure that converter, not the theoretical difference between sample rates.

G. There are some posts by Dan Lavry that may help to clear up some of the issues you were bringing up. It's confusing I know. The points that you brought up seem to be contradicting what he is saying here. But again, the difference is are we talking A/D D/A conversion or basic DSP processing.

Tom, I am stuck with my pants down on this one unless/until I can find that article so I can see if there's a different IT-based angle altogether - which I seem to remember - which would be point-in-the-process independent, or whether I'm just remembering or interpreting that paper incorrectly. Until then I'm really stuck in neutral on the topic and will defer to you the benefit of the prevailing wisdom on this issue .

But in the meantime I'll still keep my signal chain short, flip only those bits that need flipping and leave the rest alone, and work at 44.1/24. If a plug internally does this or that upsampling or floating point conversion magic, more power to it; as long as what comes out of my monitors sounds good.

G.

 

[pezking]

OK I've been holding off replying to this thread for a while, and felt it was now the time to give my input.

44.1kHz was originally chosen as sample rate for CDs and other platforms as it gave a decent nyquist rate (ca. 20kHz) and allowed a small amount of space for the anti-aliasing filter.

48kHz was used for DAT recorders to allow leeway for the use of varispeed on some tape machines. When many machines are varispeeded the sampling frequency changes proportionately and the result is to shift the first spectral repetition of the audio baseband. If the sampling frequency were reduced too far, aliased components may well become audible.

The AES-5-1998 standard allows 96kHz sampling rate as an option when the audio bandwidth exceeds 20kHz, or where relaxation of the anti-aliasing filter is desired.

As far as I can gather, the main reason for having a higher sample rate (i.e. >48kHz) is to allow the use of a shallower anti-aliasing filter which may cause fewer phase problems at higher audio frequencies.

I cite most of this information from "Sound and Recording" by Francis Rumsey and Tim McCormick.

If someone claims to hear a difference between 48kHz and 96kHz, I would ask them to participate in a series of blind tests. I would suggest that this difference is a psychological effect; and a series of blind tests would see if the claim holds water.

The truth is that most people can't even hear up to 20kHz; I've seen evidence of this in hearing tests on many subjects. There is some debate, however, if we are able to sense this HF information and if we can, whether or not it affects our listening experience.

Oh and one last thing; if you really want to mix from 96kHz wav files, and find that the strain it puts on your computer is excessive, down-sample to something reasonable (44.1kHz) but don't delete the original files, do your mixing/editing, then before producing your final mix replace the down-sampled files with the original high sample-rate files. This is known as "offline editing" and is very common within the video industry when dealing with HD video compilation and editing.

The bottom line is: Will your clients hear the difference between a track mixed from 96kHz files and a track mixed with 44.1kHz files? Probably not. Is it worth the extra data? Probably not. If it makes you happy though, be my guest.

I'm sticking with 44.1kHz, and will have half as many hard drives as you guys on 96kHz

 

[masteringhouse]

The truth is that most people can't even hear up to 20kHz; I've seen evidence of this in hearing tests on many subjects. There is some debate, however, if we are able to sense this HF information and if we can, whether or not it affects our listening experience.

Alias frequencies occur at multiples below 20kHz. The question is whether can you hear them when they are present and when they are not in a track.

Nice demo here:
http://www.dsptutor.freeuk.com/aliasing/AliasingDemo.html