44.1 kHz vs. 48 kHz - why not use the higher?

[Mr. Kelly]

Hey all,

I've been noticing on here that there are a bunch of people that say that their "standard" sessions are 24 bit 44.1 kHz. I've always used 24 bit 48 kHz sessions, then truncated down to 16 bit 44.1 kHz for mastering.

Anyone have reasons, other than file sizes, to NOT use a 48 kHz sample rate? Does the truncation yield a point of diminishing return, i.e., not produce enough of an improvement to offset the detriments of the truncation?

What session rates do you use?

[albee1952]

In theory, sample rate conversion may cause errors in the audio files. THis is the same logic that records at 88.2(instead of 96K) for a final product CD. The math to convert form 88.2K to 44.1K is simple, vs the math to convert from 96K to 44.1K. My philosophy is simple. If the end goal is audio(CD or uploads) I go 44.1K. If the end goal will be tied to video, I go 48k. In either case, I always record at 24 bit and dither the final 2mix during mastering (even if its nothing more than dither and Waves L2 for a rough CD. I bounce to disk at the original sample rate and bit depth. Any conversion or bit reduction gets done in a separate step in WaveLab(including mp3). Now, having said all this, if your ears tell you that 48K sounds better, then go with it(I have yet to have a client mention it).

[Top Jimmy]

Originally Posted by albee1952

The math to convert form 88.2K to 44.1K is simple, vs the math to convert from 96K to 44.1K.

So you had to go and whip out that chestnut huh albee? While it seems logical, it really doesn't matter.

But you are right. It's the SRC that has always been frowned upon. Avoiding it is looked upon as a step towards quality.

Historically, many SRC algorithms have been quite atrocious. While many have improved over time, you can see a visual of where some are currently here. http://src.infinitewave.ca/

And of course, Audioease's world-famous audio demonstrations. http://www.audioease.com/Pages/Barba...a4SRCTest.html

[albee1952]

I only stated my opinion No one needs to agree, and other views are always welcome(since I would not claim to be the end-all on this one). Besides......he started it

[O.G. Killa]

Originally Posted by albee1952

The math to convert form 88.2K to 44.1K is simple, vs the math to convert from 96K to 44.1K.

Just want to go back and point out this is urban legend. It is simply not true. they are both equally as easy and error free.

What effects the sound of the conversion is not the actual conversion to the new sample rate, it is the reconstruction filter at the Nyquist frequency. But this holds true for any sampling, not just sample rate conversion. The crappier the filter is, the worse it will sound.

[Park Seward]

Originally Posted by O.G. Killa

Just want to go back and point out this is urban legend. It is simply not true. they are both equally as easy and error free.

What effects the sound of the conversion is not the actual conversion to the new sample rate, it is the reconstruction filter at the Nyquist frequency. But this holds true for any sampling, not just sample rate conversion. The crappier the filter is, the worse it will sound.

"The's also why resampling at integer multiples or divisors would be less likely to distort, the math is simpler and less prone to round-off error. If we're downsampling from 96 to 48 all we need do is throw away every other sample. If we're upsampling, each new sample lies exactly halfway between the old and all we have to do is set its value to the average of the two old samples on either side of it. But if that new sample has to be inserted at .345654 of the time between two old samples and the next one is at .897667 of the time difference, and only a few of the new samples fall exactly on the same time mark as the old ones, now we got the problem in interpolation."

http://www.dvinfo.net/forum/all-thin...important.html

Sampling rate conversion is simplified if rates are integer multiples of each other.

Here are some interesting charts comparing SRC with different manufacturers:

http://src.infinitewave.ca/

[spicemix]

Originally Posted by Park Seward

"The's also why resampling at integer multiples or divisors would be less likely to distort, the math is simpler and less prone to round-off error. If we're downsampling from 96 to 48 all we need do is throw away every other sample. If we're upsampling, each new sample lies exactly halfway between the old and all we have to do is set its value to the average of the two old samples on either side of it. But if that new sample has to be inserted at .345654 of the time between two old samples and the next one is at .897667 of the time difference, and only a few of the new samples fall exactly on the same time mark as the old ones, now we got the problem in interpolation."

http://www.dvinfo.net/forum/all-thin...important.html

Sampling rate conversion is simplified if rates are integer multiples of each other.

Here are some interesting charts comparing SRC with different manufacturers:

http://src.infinitewave.ca/

You are quoting some guy who for all I know is a home theatre consumer and has no professional audio expertise whatsoever.

Go ahead and try your sample dropping approach and let us know how it sounds.

Today those factors are irrelevant in the better algos.

[O.G. Killa]

Originally Posted by Park Seward

"The's also why resampling at integer multiples or divisors would be less likely to distort, the math is simpler and less prone to round-off error. If we're downsampling from 96 to 48 all we need do is throw away every other sample. If we're upsampling, each new sample lies exactly halfway between the old and all we have to do is set its value to the average of the two old samples on either side of it. But if that new sample has to be inserted at .345654 of the time between two old samples and the next one is at .897667 of the time difference, and only a few of the new samples fall exactly on the same time mark as the old ones, now we got the problem in interpolation."

http://www.dvinfo.net/forum/all-thin...important.html

Sampling rate conversion is simplified if rates are integer multiples of each other.

Here are some interesting charts comparing SRC with different manufacturers:

http://src.infinitewave.ca/

All of the effects seen in the Infinitewave graphs are due to the filters used. they explain that in the FAQ and Help.

ABSOLUTELY NO PROGRAM EVER DIVIDES BY 2 TO DOWNSAMPLE. NONE. NOT GOING TO HAPPEN. IT IS A MYTH. This is exactly the reason why I posted about this. There are so many people that still don't understand how this actually works or have been told incorrectly and then adhere to the wrong information as though it is fact.

Do this little math problem for me please.

44100 x 160 = ?
48000 x 147 = ?

What are the answers to the equations above? Noticing anything similar about the answers? What do you think that means in terms of sample rate conversion?

[O.G. Killa]

and by the way, that DVinfo forum link is sooooo far off the mark it's scary. They guy is telling people to add DITHER when sample rate converting!?!?!?! Wow... that is scary. These guys in that link are all novice/hobbyist people. Don't put too much stock into what they are saying, because after reading through the thread pretty much all of the info they mentioned was wrong (except for the fact that DVDs use 48KHz and CDs use 44.1KHz).

[O.G. Killa]

Originally Posted by O.G. Killa

All of the effects seen in the Infinitewave graphs are due to the filters used. they explain that in the FAQ and Help.

ABSOLUTELY NO PROGRAM EVER DIVIDES BY 2 TO DOWNSAMPLE. NONE. NOT GOING TO HAPPEN. IT IS A MYTH. This is exactly the reason why I posted about this. There are so many people that still don't understand how this actually works or have been told incorrectly and then adhere to the wrong information as though it is fact.

Do this little math problem for me please.

44100 x 160 = ?
48000 x 147 = ?

What are the answers to the equations above? Noticing anything similar about the answers? What do you think that means in terms of sample rate conversion?

Here's a fun little website that might ultimately shed some light for some people on sample rate conversion "math".

http://www.mathsisfun.com/least-comm...iple-tool.html

To expand on my equations I did in the other post...let's try these.

44100 x 320 = ?
48000 x 294 = ?
88200 x 160 = ?
96000 x 147 = ?

And then do these...

44100 x 640 = ?
48000 x 588 = ?
88200 x 320 = ?
96000 x 294 = ?
176400 x 160 = ?
192000 x 147 = ?

Do you see any patterns here? And by the way, this last group of numbers is what PTHD uses. Most other DAWs (and I think LE) use the the second to last group of numbers.

Enjoy

BUT!!! none of this will actually effect the audio quality of the signal. It is the Filter used at the nyquist of the destination sampling rate that effects the fidelity. The better that filter is, the better the conversion will sound. It has nothing to do with the "math" involved in converting the numbers from one rate to the other. There will never be any "decimal points" or "rounding" or "rounding errors" or "rounding distortion" when converting sampling rates. It is an integer to integer equation and for a computer is EXTREMELY simple. And it is easier to write the code for the ONE equation to handle all the sampling rates, than to have a separate equation for each sample rate combination. With that last group of numbers, you can easily go from ANY of the chosen sampling rates to ANY OTHER chosen sampling rate with one basic equation.