CLICHE QUESTION: Do you REALLY need a higher sampling rate than 44.1KHz?

[where02190]

 

Originally posted by chris carter In terms of playback only, there is no difference between 48kHz files and 96kHz files.

 

 

There most certainly is, and if you can't hear it I would seriously suggest a trip to the ENT doctor immediately, or a new career.

[chris carter]

Originally posted by where02190 There most certainly is, and if you can't hear it I would seriously suggest a trip to the ENT doctor immediately, or a new career.

 

Actually, I just had my routine hearing check and I'm happy to report that I still have some of the best ears my audiologist has even checked Probably because I'm insanely anal about protecting my hearing so much that other people laugh at me - and genetics probably has something to do with it as well.

 

That said, the audible difference people perceive with 96kHz audio is really the converter because of the placement of the filter being much higher in the frequency spectrum, well beyond what us humans can hear. Conversly, when you listen to 48kHz, the lower quality you are hearing is that dang brick-wall limiter. The reality is that you CAN make a very good 48kHz filter, it's just unreasonably expensive to do. A number of people have in fact done just that for testing purposes and have not been able to hear any difference between 48kHz with a 'perfect' filter and 96kHz. Maybe dogs can, but us humans can't.

 

So in essence, Where, what I'm saying is that what you are hearing is a real difference, but it's the filter in the converter, not the actual resolution of the audio, that you are hearing.

 

In fact, if you upsample a 48kHz file (which in theory should contain no information above 24kHz) to 96kHz and listen back it should sound better because of the filter placement in the converter. In fact, a lot of high-end audiophile consumer stuff that wackos buy to impress their friends does just that: they upsample 44.1kHz CDs up to 88.2 before conversion; and yes they do sound better (even though no additional audio information is present because there is still nothing being reproduced above 22kHz!). Now granted, if your upsampling algorithm is introducing problems, then it might actually sound worse. But it's a fun experiement; entertaining for an afternoon at a bare minimum.

 

So with real world application:

1) 96kHz recordings don't sound better when listened to: 96kHz filters sound better - regardless of whether the original recording was made at 48 or 96.

2) processor design and programming being otherwise equal, processing at 96kHz (again, regardless of if the actual audio is 48kHz or 96kHz) will generally provide better results because the same amount of mathematical distortion is spread over twice of the frequency range and since we can't hear half of that range we only hear half of the distortion. You would think this means don't bother with 96kHz audio because it's pointless. BUT, if you are running more than one process you would want to STAY at 96kHz because of rounding error in the math. Not doing so would be like doing math on a calculator and in between each step lopping off half the decimal places (or taking your your 32 bit processing and lopping it back down to 16 bits after every step) - you will introduce more error. This is why digital mix boards that operate at 96kHz tend to sound a bit nicer than digital mix boards that operate at 48kHz. Whether the bang for the buck is worth it depends on how much money you have to spend

 

... And of course at the end of the day after we've put all this effort into making stuff sound great, people go and download mp3s anyway!

[where02190]

If your ears are so good then you need to consider a new career. 96k offers twice the resolution of 48k, twice the samples per second, and a very audible difference. While differerent converters will sound different, they will all sound improved, smoother high end, tighter bottom, clearer mids at 96 vs 48. If your brain can't perceive it, the art of audio engineering is not your gift.

[skunky_funk]

 

Originally posted by where02190 If your ears are so good then you need to consider a new career. 96k offers twice the resolution of 48k, twice the samples per second, and a very audible difference. While differerent converters will sound different, they will all sound improved, smoother high end, tighter bottom, clearer mids at 96 vs 48. If your brain can't perceive it, the art of audio engineering is not your gift.

 

 

Ok, let us go back to the premises of my first post.

 

So how come a 96KHz sampling rate performs better than a 48KHz sampling rate, despite the fact that:

 

1) Most mics' frequency response start to roll off at no more than 22KHz. Hence, any mic should not be picking up any more than 22KHz.

2) You cannot hear more than 22KHz frequencies.

3) Most high-end speakers' freq. response roll off at no more than 22KHz, and even the best ones in the market with ribbon tweeters have freq. responses up to 40KHz.

[Edge100]

 

Originally posted by skunky_funk Ok, let us go back to the premises of my first post. So how come a 96KHz sampling rate performs better than a 48KHz sampling rate, despite the fact that: 1) Most mics' frequency response start to roll off at no more than 22KHz. Hence, any mic should not be picking up any more than 22KHz. 2) You cannot hear more than 22KHz frequencies. 3) Most high-end speakers' freq. response roll off at no more than 22KHz, and even the best ones in the market with ribbon tweeters have freq. responses up to 40KHz.

 

 

As per the posts above (including my own), it has very little to do with the frequencies that are being reproduced. No, you cant hear the sounds >22kHz.

 

BUT, all A/D conversion requires an anti-aliasing filter that removes ALL frequencies above the Nyquist limit (22.05 kHz at 44.1 sampling rate). You CANNOT allow ANY frequencies above the Nyquist limit, so the converters need a VERY steep filter to eliminate those frequencies. This negatively affects the sound of the captured audio, particularly at high frequencies.

 

If, however, you move the sampling rate up to 96kHz, you move the Nyquist limit to 48kHz. Now your anti-alias filter can be MUCH gentler. You are spreading this out over a much larger number of frequencies. THAT is why 96kHz sounds better than 44.1kHz.

[skunky_funk]

 

Originally posted by Edge100 As per the posts above (including my own), it has very little to do with the frequencies that are being reproduced. No, you cant hear the sounds >22kHz. BUT, all A/D conversion requires an anti-aliasing filter that removes ALL frequencies above the Nyquist limit (22.05 kHz at 44.1 sampling rate). You CANNOT allow ANY frequencies above the Nyquist limit, so the converters need a VERY steep filter to eliminate those frequencies. This negatively affects the sound of the captured audio, particularly at high frequencies. If, however, you move the sampling rate up to 96kHz, you move the Nyquist limit to 48kHz. Now your anti-alias filter can be MUCH gentler. You are spreading this out over a much larger number of frequencies. THAT is why 96kHz sounds better than 44.1kHz.

 

 

So is it better to master with a higher sampling rate, say 192KHz? Or does it depend on the hardware/software you're gonna use?

[Edge100]

 

Originally posted by skunky_funk So is it better to master with a higher sampling rate, say 192KHz? Or does it depend on the hardware/software you're gonna use?

 

 

Its complicated (how's that for a cop-out!).

 

It depends upon the converters. Top quality Apogee or Prism converters will sound better at 44.1 than Creative Soundblaster converters at 96kHz.

 

This is because the quality of the converter itself is so MUCH better, that even with the theoretical improvement in filtering on the Soundblaster does not make up for the quality of the A/D algorithms in the Apogee converters.

 

FOR THE SAME CONVERTERS, 96kHz (or 88.2) will be better. In fact, as was mentioned, 88.2 might be better (or 176.4) since it will be reduced to 44.1 for CD.

 

But as with anything, its impossible to say whether 96kHz on one converter will be better than 44.1 on another. You need to hear them to know. But, in general, on the same set of converters, 96kHz should be better due to better anti-alias filtering.

 

Software shouldnt affect anything.

[Matt Hepworth]

The anti-aliasing comments are definitely valid, and I had disregarded that in my previous posts.

 

But, with a good converter even 44.1kHz (or at most 48kHz) anti-aliasing should be a complete non-issue. Also, let's keep in mind that there will be little to no audio information above 20kHz to be filtered anyway as the audio has already passed through several stages of 20Hz to 20kHz frequency response mechanical filtering prior to conversion. I still maintain that some converters will definitely sound better at 44.1kHz than they do at 96kHz. Perhaps it's a "good noise" like dither...

 

The whole theory about 88.2kHz and 176.4kHz makes sense, but are there not very few algorithms that actually use even math?

 

And hey, there is one tangible benefit to higher sampling rates - lower latency.

[Edge100]

Originally posted by Matt Hepworth The anti-aliasing comments are definitely valid, and I had disregarded that in my previous posts. But, with a good converter even 44.1kHz (or at most 48kHz) anti-aliasing should be a complete non-issue. Also, let's keep in mind that there will be little to no audio information above 20kHz to be filtered anyway as the audio has already passed through several stages of 20Hz to 20kHz frequency response mechanical filtering prior to conversion. I still maintain that some converters will definitely sound better at 44.1kHz than they do at 96kHz. Perhaps it's a "good noise" like dither... The whole theory about 88.2kHz and 176.4kHz makes sense, but are there not very few algorithms that actually use even math? And hey, there is one tangible benefit to higher sampling rates - lower latency.

 

There are a number of concerns with what you are saying regarding filtering.

 

The primary concern is that that is not how anti-aliasing filters work. Remember that ALL information above 22.1 kHz MUST be removed. For this reason, the filter must start removing LOWER THAN the Nyquist limit. Even a 10-pole filter (60dB/octave) would need to start filtering at well below 22.1 kHz to have those frequencies above 22.1 kHz effectively removed from the A/D stage. No filter is infinitely able to remove those frequencies, so for that reason they must start filtering BEFORE the Nyquist frequency.

 

In a 96kHz system, you could START filtering at ABOVE the Nyquist limit, AND use a much softer filter, thereby allowing those high end frequencies from appearing un-filtered (or VERY mildly filtered).

 

I do, however, agree with your statement about latency.

[Matt Hepworth]

Right, but with so little content at 20kHz and beyond anyway, plus with the frequency reproduction on most equipment not being able to do much above 20k, and with so few able to hear anything much above 18k, with a well designed FIR filter in the converter, there should not be an audible difference between any sampling rates from 44.1kHz on, correct?

 

That said, I definitley see why anti-aliasing at 48kHz by converting at 96kHz would be better on paper, but in practical use it should be of no benefit.

 

To paraphrase Roger Nichols, if you downsample from 96kHz to 44.1kHz you'd be using the same steep filters to cut off all frequencies above 22kHz as if you'd just recorded to 44.1kHz in the first place. At some point, the anti-aliasing filter is happening to content below 22kHz.

 

Admittedly, this topic has caused me to really stop and think about my stand on the issue, and I'm quite sure I'll be trying 96kHz again shortly to see if I'll hold my ground, or eat my words.

 

However, my original point was and still holds true, it's all about the converter, as there is ZERO fundamental difference from 20Hz to 22kHz between 44.1 kilosamples per second and 96 kilosamples per second. I will yield as to the anti-aliasing comments, as I see how they could be of benefit to record at higher sampling rates.

 

Thank you for replying Edge and Chris!

[Matt Hepworth]

Oh yeah, here's a great link: http://recforums.prosoundweb.com/index.php/t/2260/0

[Edge100]

 

Originally posted by Matt Hepworth Oh yeah, here's a great link: http://recforums.prosoundweb.com/index.php/t/2260/0

 

 

Thanks!

 

Great thread, BTW. Nice to has this stuff out and help us all make better recording choices!

[Toxicpotatoe]

 

Originally posted by Matt Hepworth ...with a well designed FIR filter in the converter, there should not be an audible difference between any sampling rates from 44.1kHz on, correct?

 

 

I think there are limitations on how well a filter can perform in real time, even in the digital domain.