Topic: 48k Vs. 96k..?

[ideal77dlr]

Is there a real noticable differnce between recording at 48k and 96k? Espeically if you're recording at 24bit?

I understand from an archival point of view, you've got a higher sample rate at 96k so it's archivable and usable as far a future formats are concerned, but if you're going to downsample to put on a cd, are you going to end up with a 'lesser' quality recording to start with if you run 24/48 when you go out and do your recording?
I'm thinking as well as what you save in terms of card life and potential 'slow card' error or wotnot that you hear about running at 96k...

[rhinowing]

well, i wouldn't record at 24/44.1 because of sampling stuff

as I recall, when resampling, a recording with a clean sample rate/bitdepth ratio gives a cleaner resample/dither

so I'd probably recording 24/48 instead of 44.1

i don't have a 24/96 recorder, so I can't answer the main question

[ideal77dlr]

Ah...and I did mean 44 

Edited to clarify!!!

[runonce]

A "noticeable" difference. I doubt it.

The thing you might notice is the huge file sizes.

[Brian Skalinder]

as I recall, when resampling, a recording with a clean sample rate/bitdepth ratio gives a cleaner resample/dither

You've lost me.    Please explain?

[rhinowing]

as I recall, when resampling, a recording with a clean sample rate/bitdepth ratio gives a cleaner resample/dither

You've lost me.    Please explain?

sorry--that wasn't very clear (what I get for writing posts in the middle of an all-nighter)

I'm not even entirely sure about this (I'm sure someone like DSatz could explain better), but the resampling/dithering algorithms (r8brain's, for example) perform better with a file where the sample rate/bit depth are multiples of each other (eg 24/48, 16/32, 32/96) than on one where they aren't (16/44.1, 24/44.1). I might be getting this mixed up with just sample rate, though (for example, resampling 96 > 48 or 88.2 > 44.1)

[jlykos]

as I recall, when resampling, a recording with a clean sample rate/bitdepth ratio gives a cleaner resample/dither

You've lost me.    Please explain?

sorry--that wasn't very clear (what I get for writing posts in the middle of an all-nighter)

I'm not even entirely sure about this (I'm sure someone like DSatz could explain better), but the resampling/dithering algorithms (r8brain's, for example) perform better with a file where the sample rate/bit depth are multiples of each other (eg 24/48, 16/32, 32/96) than on one where they aren't (16/44.1, 24/44.1). I might be getting this mixed up with just sample rate, though (for example, resampling 96 > 48 or 88.2 > 44.1)

IIRC, it's the resampling that is most affected if you go in a non-multiple, like 24/96 > 16/44.1.  The bit rate conversion is generally OK (you can't get around it anyway if you convert from 24-bit to 16-bit), but if you record in 24 bit and primarily listen at 16 bit, it is generally preferable to record at something like 24/44.1 or 24/88.2 because they are clean multiples of 44.1.

In a program like Sound Forge or Audition, there are different noise shaping algorithms that one can use to mitigate the negative sonic effects of resampling.  I record in 24/96 and then use triangular noise shaping to resample to 44.1.

[taperj]

well, i wouldn't record at 24/44.1 because of sampling stuff

as I recall, when resampling, a recording with a clean sample rate/bitdepth ratio gives a cleaner resample/dither

so I'd probably recording 24/48 instead of 44.1

i don't have a 24/96 recorder, so I can't answer the main question

Actually, if you're trying to end up with a cd quality release 24/44.1could be considered better to use than 48 or 96 because of the reason you specified. Going from 24/44.1 > 16/44.1 there would be no resampling which would make it the cleanest since all it would need to do is drop the last 8 bits to get to cd quality. Resampling from 96 or 48 to 44.1 is where the mathematical/interpolation conundrum comes in. Logically for high res recording 88.2kHz would be the next best since it is evenly divisible by 2 which shouldn't pose a mathematical problem when going to 16/44.1. As far as your 96kHz question, yes, I do notice a difference between 48 and 96 kHz. It's a bit of an intangible but really I don't hear the difference, I feel it. The weight of how the music hits and effects my body and environment is different and somehow heavier and richer. I think this mostly comes into play on a good rendering system played loud. Other than pure end point sound quality there are other reasons to run at higher resolution, one being the greater ability to work with the recording in post. you simply have more to tweak with a 24/96 recording. I think this is a good/fairly simple read about some of the why's/why nots:

http://www.tweakheadz.com/16_vs_24_bit_audio.htm

J

[phanophish]

http://www.korg.com/services/products/mr/Future_Proof_Recording_Explained.pdf

I find this article does a pretty good job explaining the limitations of the various PCM sample rates/depths.  It is a Korg marketing piece so filter the data a bit.

That said I usually run 16/44.1 on my R44.  The file sizes of 24/96 and post processing work to render a 16/44.1 copy are not worth the incremental improvement in sound quality that you get with 24/96. 

[Belexes]

That said I usually run 16/44.1 on my R44.  The file sizes of 24/96 and post processing work to render a 16/44.1 copy are not worth the incremental improvement in sound quality that you get with 24/96.

To each their own, but I like the extra headroom 24 bit gives.  My work in post for the files is simply to import into Audacity and export to 16/44.  There really isn't that much work to it.

I tape 24/48 all the time. never saw the need to jump up to 96.

[phanophish]

I get the advantages and headroom of running 24 bit instead of 16, I'm familiar enough with my gear that I can usually run the R44 hot enough at 16 bit that I don't feel the need for the extra headroom.  If I'm taping in a room I don't know as well I will sometimes bump up to a higher bit depth.  But why 48 instead of 44.1?  My understanding is the same as JLYKOS in that running a non multiple of 44.1 for the sampling rate makes the re-sample "less clean".  It makes you much more dependent on the quality of the re-sample algorithm rather than the quality of the actual recording.  All this presumes that the ultimate goal here is a 16/44 CD.  Obviously there is a slight quality bump if you are playing back the 48k files without changes.

[boojum]

I always record at 24 bit depth.  I like the greater dynamic range.

And I usually have a sample rate of 44.1.  I cannot understand why there has to be any connection between bit rate and sampling rate.  If there is any published literature on this I would like to read it.  As for sampling rates, well, this is like talking about the events in the US between 1860 and the present.  There are strong opinions on both sides of the issue, lots has been written, but I do not know any definitive proof over which is the best rate.  I ran a pull for a fellow who is a real pro and he insisted on 24/882.2.  OK.  I am not sure how many folks could distinguish sampling rates in a double blind test.  When they can be distinguished with a high rate of accuracy I will use that rate.

What is important is where those mics are.


Cheers

[Tim]

IIRC, it's the resampling that is most affected if you go in a non-multiple, like 24/96 > 16/44.1.  The bit rate conversion is generally OK (you can't get around it anyway if you convert from 24-bit to 16-bit), but if you record in 24 bit and primarily listen at 16 bit, it is generally preferable to record at something like 24/44.1 or 24/88.2 because they are clean multiples of 44.1.

In a program like Sound Forge or Audition, there are different noise shaping algorithms that one can use to mitigate the negative sonic effects of resampling.  I record in 24/96 and then use triangular noise shaping to resample to 44.1.

^^^ this best fits my understanding as well.

[jeromejello]

most of the time i go 24/44.1
some of the time i go 24/48

never thought to do anything but based on both of my current needs [audio (44.1) / video (48)].

have been thinking of going all 48, but it hasnt really stuck.

[Belexes]

All this presumes that the ultimate goal here is a 16/44 CD.  Obviously there is a slight quality bump if you are playing back the 48k files without changes.

Hopefully one day we will get away from 16/44 since the CD is becoming a thing of the past.

I do play a lot of my masters from my computer with headphones on and don't burn to CDR, thus I do a lot of playback 24/48.

[rowjimmytour]

When I moved up to 24 bit I started at 24/48 and then read about the sample rate ratio deal I think from dsatz or another member and went to 24/44.1. I was told the only reason to tape 24/48 is if you want to sync video w/ your recording to run w/ programs like vegas.

[anr]

Just like to say many thanks for the informative replies.  It answered a problem I had.  I was asked to edit and transfer a 48/16 recording to CD (Acoustic guitar and vocal).    I used dBPoweramp and the result was horrible - a nasty "ringing" in the acoustic guitar.   I realised the problem was in the conversion but not why.  I subsequently did the conversion in Soundforge and the ringing has disappeared entirely.   

[phanophish]

When I moved up to 24 bit I started at 24/48 and then read about the sample rate ratio deal I think from dsatz or another member and went to 24/44.1. I was told the only reason to tape 24/48 is if you want to sync video w/ your recording to run w/ programs like vegas.

FWIW - I have used 16/44.1 audio in Vegas without issues though.  Obviously it is converted but the quality was certainly acceptable.  I could not hear a difference in casual listening.

[rowjimmytour]

When I moved up to 24 bit I started at 24/48 and then read about the sample rate ratio deal I think from dsatz or another member and went to 24/44.1. I was told the only reason to tape 24/48 is if you want to sync video w/ your recording to run w/ programs like vegas.

FWIW - I have used 16/44.1 audio in Vegas without issues though.  Obviously it is converted but the quality was certainly acceptable.  I could not hear a difference in casual listening.

I might be out of my league here about video combined w/ audio but I think the point of 24/48 was to have the highest quality audio for a video so no conversion would take place

[bgalizio]

Also, IIRC, DVD-A can do 24/44, 24/48, 24/88, 24/96 etc. But, creating an audio only DVD-V disc requires 24/48 or 24/96 etc. Some people want to listen to 24bit audio via a DVD-V disc and cannot play DVD-A discs.

[anonymous_user]

I might be out of my league here about video combined w/ audio but I think the point of 24/48 was to have the highest quality audio for a video so no conversion would take place

I think the reason that was stated for video projects is for the same reason you would otherwise record in 24/44.1, because it prevents resampling, or specifically resampling from 44.1 to 48 or vice versa. The audio sample rate used for a standard DVD is 48 or 96 so the final audio needs to be in that format.

[phanophish]

I might be out of my league here about video combined w/ audio but I think the point of 24/48 was to have the highest quality audio for a video so no conversion would take place

I think the reason that was stated for video projects is for the same reason you would otherwise record in 24/44.1, because it prevents resampling, or specifically resampling from 44.1 to 48 or vice versa. The audio sample rate used for a standard DVD is 48 or 96 so the final audio needs to be in that format.

I guess the issue is rarely do I record specifically for video.  So it becomes a question of which end product do I want to have a resample in the audio post processing path.  For me I would rather have that on the video gen and keep the audio as intact as possible

[M]

Quantization errors anyone?

We all know that there are only two samples per cycle required to recreate a waveform.  If your cut off frequency is close to the range of signal that is important than this will cause errors.

What happens is just above the cutoff frequency the audio is being sampled less then two times per cycle and this is the cause of errors.

I believe that there should be adequate room for the high frequencies that will be misinterpreted to be rolled off before being sampled.  My perfect PCM system would sample at 96khz but the rolloff of high frequencies will start at 60kHz. 

[page]

I sample at 44.1 for two main reasons: I trust my SD box more at 44.1 then I do my cheap wav editor's routine to take it's 96k signal and resample it to 44.1. Second, unlike my pets, my hearing maxes out somewhere above 17khz anyway. At the end of the day, I'm not using DVD-V (which as has been noted requires certain frequencies) so I don't gain anything by burning the space. I figure there are other things that will cause problems with the sound and I work on those first. ymmv.

So to answer the OP question, of the two, I'd go for 48k. best of luck.

[DSatz]

Ideal77, if you'd asked whether a particular recorder performs better at one sampling rate than another, you might have gotten a clearer answer. But since you asked about the difference in general, as you can see, the discussion tends to become quasi-theoretical. Since so many myths about digital audio are still in active circulation (several of which have already appeared in this thread), the answers that you get (a) won't usually agree and (b) even if they do agree, might be flat wrong, and (c) even if they agree and are scientifically accurate, still might not help you very much, if you're just trying to decide what sampling rate to use for your next recording, or which recorder to buy.

One thing I can simplify for you right away, though: You mentioned the number of bits per sample. That number, whether it's 16, 24 or 3.14159, has nothing to do with sampling rate. An oversupply of one can't help an undersupply of the other. The number of bits per sample, if it affects anything at all, affects ONLY the dynamic range--the maximum difference between the highest possible sound level and the noise floor of the recording. It can't affect anything else unless the design of a system is hopelessly broken in some way.

It's not like with MP3 files, where there's a single "bit rate" (such as 96 kbps) and the higher that bit rate, the better the sound quality is likely to be (up to a point, anyway). Instead, there are two separate specifications--sampling rate on the one hand, and "word length" or "bit depth" on the other--each of which influences a different, mutually exclusive, dimension of sound quality.

--best regards

[nusymphony]

stick with 48k. Less space, less hassle and you probably won't be able to hear the difference

[DSatz]

A/D conversion seems to remain one of the more frequently misunderstood aspects of recording. I think one of the problems may be that some parts of the process are easier to visualize than others, and that people focus more on the parts of the process that they can readily visualize, even though those parts aren't enough to clarify the A/D - D/A process as a whole.

I hope I can challenge some of the statements in this thread without offending anyone. As time allows, I'd like to go over some of the ideas that have been put forward here which are either largely or completely mythical. I'll put them into my own words because I'd like to show respect for the individuals who have posted these assertions in good faith. Here's the first one:

"When converting a recording from one sampling rate to another, you'll get better performance and/or better results if you convert between rates that have a simple ratio, e.g. 96 kHz to 48 kHz or 88.2 kHz to 44.1."

This actually has two independent claims in it: "The process will run faster if ..." and "The process will produce more accurate results if ...". Neither claim is correct as a rule.

I think I read that someone here had written his own software routine to resample a WAV file down to half its original sampling rate. That's admirable if so. It would also be just about the only situation in which the "run faster" part of the above statement might be true, since the whole sampling rate conversion (SRC) procedure is then designed around a special case. Unfortunately, the software would then be completely unable to handle any other conversions than the case that it was designed for.

99.999% of the time, people use software (or in many professional applications, hardware) that can convert between any two chosen sampling rates. In live recording and broadcasting, the input and output sampling rates might not even be known precisely in advance, and/or their relationship may not be entirely stable. The SRC device has to track both rates independently, and provide accurate conversion while its inputs and its outputs are "synched" to two different clocks that can vary in real time, without adding jitter. That's a neat trick, but it can be done and is done every day nowadays with no audible loss in quality.

Because of this requirement for generality--even the simple ability to specify any two known, fixed rates as in most music editing software--the optimizations which might/could have been used for the special case of simple ratios can't be used. Thus with any commonly available SRC software and hardware, there is no advantage to choosing simple ratios for conversion.

There is also no general increase in numerical accuracy when simpler ratios are used, but to explain why would require another message, and now it's time for breakfast, so maybe later.

--best regards

[Shadow_7]

Record in the highest sampling rate you can afford.  HDDs are cheap.  Audio is cheap (in relation to other media types).

For video you might have differences in the clock between the audio recorder and the video recorder.  You don't want to use the video recorders audio for anything other than syncing.  Since it's encoded to AAC, MP3, or any number of hideous incantations.  So most times you're not only resampling, but compressing or expanding the length of time it runs in to match up with the video.  Basically if you're doing a lot of EDITing, you NEED the higher rates.  Extra bits 16->24->DSD gives you a greater dynamic range, which is more of a safety feature and built in compression of sorts for dynamic range when converted.  I prefer to edit at 24/96 because it loads in audacity in a reasonable amount of time.  And I can apply EQ and other edits and still render out to 16/48 in realtime to play on my 16/48 soundcard on my 4yo laptop.  If you're crunched for time and can't live in excess, then probably not for everyone.

I record in DSD, and convert to 32/192 in audiogate.  Used to be 24/96.  Then I use sox to resample, apply blatant gain, and trim as needed.  From that version I do my edits.  Which might be a little hard limiting, maxed out gain / normalization / amplification.  And convert to a deliverable.  But I rarely deal with CDs as anything other than a distribution format.

[M]

Shadow

I believe in PCM, sampling above 96kHz actually degrades the signal.  I think that the converters cannot encode/operate that fast and the signal ends up being less accurate than if it were recorded at a lower rate.   

Dan Lavry Sampling Theory for details:

http://www.lavryengineering.com/documents/Sampling_Theory.pdf

DSD is another story and and it uses high sampling rates for different reasons than PCM.