Flowchart for selecting the correct sample rate

I've made this flowchart which could be useful when producing or recording.

 

There are cases where upsampling or playback and capturing at different sample rates makes sense (some mastering engineers do this), but not during production.

 

http://www.centerforlydteknik.dk/download/pdf/sample-rate-flowchart.jpg

 

In case you need to use existing libraries or audio files with different sample rates, you should upsample the lower files to the highest common sample rate using an offline converter or your DAW's built-in offline import converter. Be aware that inter-sample peaks can manifest themselves during such conversions, which can lead to a loss of headroom and audible distortion.

 

Logic Pro can handle realtime sample rate conversion in the EXS24 sampler, but this produces inferior results.

Edited February 10, 2014 by lagerfeldt

Cool, I'm sure lots of people will find this helpful.

 

Stickie it I say!

Thanks.

 

Allow me to quote myself from a related post on Gearslutz:

 

One of the main problems with using e.g. 44.1 kHz sample libraries (such as most drum libraries) in a higher sample rate project is the inevitable low quality realtime sample rate conversion and manifested inter-sample peaks from heavily processed samples which will severely degrade the audio quality. This will by far offset any filtering advantages of higher samples rates.

 

Even if the entire library is offline upsampled to the higher project sample rate the manifested inter-sample peaks and some SRC degradation would still be present. Lupo wrote this nice article about ISP's:

http://www.gearslutz.com/board/showwiki.php?title=Tips-and-Techniques:Intersample-peaks

 

As for using >96 kHz sample rates Dan Lavry recently wrote this article:

http://www.lavryengineering.com/pdfs/lavry-white-paper-the_optimal_sample_rate_for_quality_audio.pdf

Thanks for this, Lagerfeldt. Your posts have always been a great help to me (I loved your write up on ducking).

 

According to this Apple Support document, "higher sample rates result in less latency at the same buffer setting. For example, if you set a buffer size of 128 samples, the latency added by the I/O buffer with a sample rate of 88.2 kHz will be half as much as with a sample rate of 44.1 kHz." Source: http://support.apple.com/kb/HT1314

 

For those of us who track timing sensitive vocals/instruments, what are your thoughts on selecting an ideal rate for timing issues as opposed to strictly sound quality issues? Is increasing the sample rate to achieve lower latency simply the wrong approach (wrong, in the sense that it will lead to the problems you've convincingly detailed later on)?

 

You're very generous with sharing your time and experience, so, to be clear: I'm not asking you to get into a discussion about latency, per se. Just sincerely interested in your thoughts on this specific aspect of things.

I'm glad you find my posts helpful.

 

Higher sample rates do result in less latency at the same buffer setting. Higher sample rates also put more stress on the system and may require a higher buffer setting to avoid instability or drop-outs, i.e. it could be self defeating.

 

If your system can handle the extra stress, then a higher sample rate will help with your latency. However, since the latency problem is usually a monitoring issue, the solution is not to choose your sample rate primarily based on latency, but to use a direct monitoring system that avoids the AD-DAW-DA roundtrip. Many professional sound cards offer some way of direct monitoring, though you may have to live with AD/DA latency depending on where and how the card taps the I/O's.

 

Specs for converter latency in the Lavry Blue:

 

At 44.1 kHz:

AD converter latency is 1.375 msec

DA converter latency is 1.69 msec

 

At 48 kHz:

AD converter latency is 1.252 msec

DA converter latency is 1.535 msec

 

At 88.2 kHz:

AD converter latency is 0.688 msec

DA converter latency is 1.08 msec

 

At 96 kHz:

AD converter latency is 0.634 msec

DA converter latency is 0.993 msec

Edited May 12, 2012 by lagerfeldt

Awesome. Thanks for clearing this up for me

+1 Thanks to Lagerfeldt. Can't thank you enough for taking so much time to share your valuable knowledge here. This chart is so helpful. Thank you!

Thanks a lot Lagerfeldt.

 

Could you tell why does 192kHz degrade audio quality? I'm curious.

There's nothing theoretically wrong with using such a high sample rate. It's superfluous and much harder on the CPU, but not wrong from a theoretical standpoint.

 

However, the actual physical components in an analog to digital converter cannot use such a high frequency rate without degrading the captured signal. So at 192 kHz the captured signal is a lot less than 24 bit resolution, but closer to a 16 bit signal.

 

Think of it like this: a high speed camera may be able to capture a large number of snapshots per second, but the quality of each picture is degraded due to the very short exposure time.

 

Dan Lavry explains it in his PDF: http://www.lavryengineering.com/pdfs/lavry-white-paper-the_optimal_sample_rate_for_quality_audio.pdf

 

Why was 192 kHz invented? Marketing bullshit that preyed on some people's insecurities or misunderstanding of digital audio, and it became yet another tickbox - just because it was technically possible. But you won't find it on any Lavry converter.

However, the actual physical components in an analog to digital converter cannot use such a high frequency rate without degrading the captured signal. So at 192 kHz the captured signal is a lot less than 24 bit resolution, but closer to a 16 bit signal.

There's also the issue of clock jitter: a clock running at 192kHz will be less precise than one running at 48kHz, introducing more timing errors which help further degrade the signal upon conversion.

That is indeed part of what I am referring to, but it is related to the physical properties of the design, not any theoretical limitation in clock design.

 

In any case, 192 kHz makes no sense since it is beyond even the maximum necessary to avoid any audible alias filtering, which only underscores the marketing b.s.

Great thread and very helpful chart. But what if your project is intended for CD? Is the "cost" of sample rate conversion worth the additional quality of using a higher sample rate? If yes, which one yields the cleanest down conversion to 44.1? What do you recommend in this case?

Great thread and very helpful chart. But what if your project is intended for CD?

You should still follow the chart.

 

Is the "cost" of sample rate conversion worth the additional quality of using a higher sample rate?

It depends, but theoretically, yes. If you record multiple analog sources at 48 kHz, process and bounce to a stereo 48 kHz file you will have benefited from the higher sample rate on multiple channels, but only had to live with one stage of (hopefully high quality) sample rate conversion.

 

If yes, which one yields the cleanest down conversion to 44.1? What do you recommend in this case?

Izotope RX gives you the best quality and most flexible parameters.

 

However, I recommend you stay at the original sample rate (e.g. 48 kHz) and let your mastering engineer perform mastering at the original rate. Then let him convert to your destination rate, e.g. 44.1 kHz.

 

If you perform DIY mastering be wary of inter-sample peaks on SRC'ed clipped or limited materia, and leave some extra headroom.

Just to clarify, are you suggesting 48 kHz, or would you still recommend 96 kHz for a CD project if the flow chart leads there (which it does in my case)?

Follow the flowchart, which leads to 96 kHz in your case.

 

Contrary to popular belief there is no major advantage to recording in 88.2 kHz over 96 kHz if your project will end in 44.1 kHz. The myth goes that sample rate conversion from 88.2 kHz to 44.1 kHz is much better because the numbers are even multiples. However, most offline sample rate converters work by first finding the lowest common multiple and going back from there.

 

The exception being when you are applying realtime SRC on some signals due to mixed sample rates. This can happen in live productions or in a studio production with samplers/ROMplers using unmatched sample rates. In that case there are benefits to using easily divisible sample rates together, but the overall quality is worse than offline SRC and it should be avoided.

Thanks Lagerfeldt, I really appreciate your expertise. One more question if you don't mind: in the above post, you recommend offline sample rate conversion (SRC) as part of mastering, using iZotope RX. How does Logic's built-in SRC (part of the Bounce command) compare with offline conversion using iZotope RX? If you don't have RX or a similar offline SRC available, would that change your opinion on recording at a higher sample rate and then down converting? Edited January 2, 2013 by ERO

It's worse. Whether you will be able to hear it or not depends on the material. You can check out various algorithms here: http://src.infinitewave.ca/ Please notice that Izotope RX2 includes the 64 bit algorithms as well.

 

If you don't have RX or a similar offline SRC available, would that change your opinion on recording at a higher sample rate and then down converting?

That depends on what is available to you, Barbabatch does decent SRC for instance, but not on par with Izotope. Logic's SRC is not that bad, but I wouldn't use it for mastering.

If Izotope RX is out of your budget, you can get the sample rate conversion for $60 by purchasing Triumph.

If Izotope RX is out of your budget, you can get the sample rate conversion for $60 by purchasing Triumph.

 

Triumph looks great for sample rate conversion and mastering on the cheap, by using the iZotope engine (64-bit SRC and MBIT+ dither). I need to convert and master for a small project. Is Triumph too good to be true for $60?

 

David: Have you used Triumph to master anything? Pros/Cons? Anyone else?

I haven't used Triumph yet, but I rarely do any mastering myself and if I do, I try to avoid SRC as much as possible.

Lagerfeldt, have you used any of the new Apple Mastering Tools? Curious to know what you think of the SRC in the afconvert tool? Do you know if this is the same SRC used by Logic Pro?

I believe it's all based on Core Audio. I only use the RoundTripAAC and afclip command-line tool.