Topic: Resample to 44.1 or record at 44.1 (midside Q)

[page]

I've been playing around with my recorder when doing MS (raw encoded) stuff, and I've tried doing 88.2 resampled to 44.1 in Audacity and recording straight to 44.1.

From a technical midside decoding standpoint, is there an advantage to doing the resample, or is it a bust and recording at 44.1 is good enough? Does it make a difference in the end product (at 16/44.1)?

I haven't had a chance to do any real comps in a controlled environment, so I don't know. I figure someone out there has looked into this before.

[boojum]

Sampling rate discussions can go on as long as the pious could argue during the Middle Ages about how many angels could dance on the head of a pin.  People who really know (Lavry) have argued and written at length that above 48kHz is kind of a waste.  I may be wrong about the exact number, but the fellow argues that filters and so on make really high sampling rates ridiculous.  I also wonder if binary math is easier on numbers which are multiples of two in decimal; i.e., 88.2 to 44.1  That said, I record at 24/48.  I know some folks say they can hear the difference between 48 and 96 and 192.  I have yet to see one person do this in a double blind test (ABX).  If it has been done, please show me the test results. 

For MS I cannot imagine why rates would be any different than other mic arrays.  I use MS a lot.  Love it in clubs, etc., where I have to set up close and in a hurry.  It is also inconspicuous which makes the performers and management happy.

As usual, YMMV     

[page]

Sampling rate discussions can go on as long as the pious could argue during the Middle Ages about how many angels could dance on the head of a pin.  People who really know (Lavry) have argued and written at length that above 48kHz is kind of a waste.  I may be wrong about the exact number, but the fellow argues that filters and so on make really high sampling rates ridiculous.  I also wonder if binary math is easier on numbers which are multiples of two in decimal; i.e., 88.2 to 44.1  That said, I record at 24/48.  I know some folks say they can hear the difference between 48 and 96 and 192.  I have yet to see one person do this in a double blind test (ABX).  If it has been done, please show me the test results. 

For MS I cannot imagine why rates would be any different than other mic arrays.  I use MS a lot.  Love it in clubs, etc., where I have to set up close and in a hurry.  It is also inconspicuous which makes the performers and management happy.

As usual, YMMV     

Yeah, I can't hear much difference between 44.1 and 48, and I can't tell a difference between 48 and 96 at all, however from a math and processing standpoint, I'm curious if it affects the final result of the MS decode, and thus mixing at 88.2 then downsampling would produce a different result then just mixing at 44.1. I have a tendancy to agree, in that I don't think it does, but I'm curious.

[DSatz]

boojum, you wrote:

> I also wonder if binary math is easier on numbers which are multiples of two in decimal; i.e., 88.2 to 44.1

Some SRC routines run more efficiently with simple integer ratios while others don't, so this depends on what software you have. I ran my own little test just now: I took 3 minutes of music recorded at 96 kHz, and converted it both to 48 kHz and to 44.1 kHz. With Adobe Audition 1.5 the conversion to 48 kHz took about half as much time as the conversion to 44.1, but with Sound Forge 7.0 the times required for the same two conversions were within 1% of each other. So, let's beware of generalizations that are overly broad.

Precision and accuracy are what could affect the sound quality of the result, and they're a separate issue in the design of an SRC routine. (They shouldn't be too hard to measure, but I have yet to see any data at all about that in any software reviews.) In the end, I'd say that whether the eventual 44.1 kHz recording might sound better if you start out recording at some higher sampling frequency depends on the equipment that you record with originally, on the particular SRC method that you use, and on whether more accurate audio pleases your particular ears. There can't possibly be one answer that everyone would always agree with.

Technically there's no reason why 44.1 kHz can't sound as good as higher sampling rates, but that is not to say that all existing 44.1 kHz converters sound good to everyone. Whatever equipment you're working with--regardless of what might have been possible in theory if things had been done differently--you surely want to use it in the way that you think sounds best. There's something to be said for the "let the Wookiee win" principle.

--best regards

[digifish_music]

I've been playing around with my recorder when doing MS (raw encoded) stuff, and I've tried doing 88.2 resampled to 44.1 in Audacity and recording straight to 44.1.

From a technical midside decoding standpoint, is there an advantage to doing the resample, or is it a bust and recording at 44.1 is good enough? Does it make a difference in the end product (at 16/44.1)?

I haven't had a chance to do any real comps in a controlled environment, so I don't know. I figure someone out there has looked into this before.

I take a pragmatic approach to this question (as I have failed to hear any difference in A/B blind listening).

If I am working on an audio project it is 44.1. If I am working on audio for video then 48 kHz. If I am recording samples with the view to transposing them downward (and so care about ultrasonic frequencies) I try to work at 2X the final sample rate (88.2 or 96 kHz).

digifish

[run_run_run]

I don't get the 2 times logic, please fill me in.

[rhinowing]

I don't get the 2 times logic, please fill me in.

nyquist--you have to sample at twice the frequency of your signal in order to avoid aliasing

[Lil Kim Jong-Il]

An important consideration is target format for the 24-bit product.  A DVD-A will carry 44.1k or 48k  but a DVD-V format "audio DVD" is restricts LPCM streams to 48k or 96k.  Not all DVD players will read DVD-A format; DVD-V format is universally compatible.   

If you plan to circulate 24-bit discs, you should record at 48k so that you have the option to author into both formats. 

[Lil Kim Jong-Il]

I don't get the 2 times logic, please fill me in.

nyquist--you have to sample at twice the frequency of your signal in order to avoid aliasing

I think he's talking about twice the sampling frequency for the working bit stream, not twice the highest frequency he intends to reproduce.  88.2 is 4X the highest audible frequency one would expect to reproduce from a CD. 

The usual argument for processing at the higher rates and then down sampling is that the digital processing (filters, EQ, etc) produce better results when working with more data.  I don't have any experience with post processing tools so I don't know if any difference in product is audible.

[run_run_run]

I think when I do 24 bit in safe venues with my laptop, I am going to run 24/48 from now on. I normally go 24/44.

[DSatz]

When you know you're going to release something in a 16-bit format, it makes complete sense to record 24-bit, and do all your processing with that increased level of precision before you set your final levels and dither the signal down to 16 bits.

However, it makes no such sense to record at a higher sampling rate (sampling frequency) unless you have some evidence that your particular signal processing chain works better that way--a conclusion which would require careful testing. In general it shouldn't be expected; there's certainly no theoretical reason to expect it. Sampling rate and bit depth are entirely different things.

It could be an advantage in a particular case, due to some shortcoming of a particular 44.1 kHz converter, for example. But if that were a widespread situation, then recording at 96 followed by conversion to 44.1 ought to sound distinctly better to the people who prefer 96 kHz recording than recording at 44.1 in the first place sounds. In general that doesn't seem to be true.

Think this through with me, please: When you sample at (say) 96 kHz, the signal components below 22.05 kHz aren't recorded any more accurately than they are recorded by sampling at 44.1 kHz. But the filtering occurs at 48 kHz instead of at 22.05, and even though no one can hear between 22.05 and 48 kHz, the signal components below 22.05 kHz may be handled in a more linear fashion. Such is the rationale for 96 kHz recording in a nutshell.

Unfortunately, when you then convert down to 44.1 for release, you must filter a second time (at 22.05 kHz) to avoid aliasing distortion. Signal components between 22.05 and 48 kHz can't be allowed at that point. So in the end, you put the signal through more filtering--not less--by starting out at a higher sampling rate. The entire approach is self-defeating in that sense.

--best regards

[SmokinJoe]

If you seriously think that now, or someday, you may want to playback on a format where the required bitrates are 48k or 96k, then record at that rate, and resample to 44.1k for CDs.  But if you don't see that happening, or can't be bothered with backing up those original files, then you are better off to record at 44.1k

I think it is a huge misconception that people have that recording at 48k, 96k, or 192k and then resampling to 44.1k in post is superior to an original recording at 44.1k.  That's not true.  It can't be.  Your A/D oversamples the signal by a large factor, and it has lots of information in front of it to make good decisions, and it is MUCH better suited to make these decisions than your computer.  Let it do it's job!  In post your computer has much less information to go buy, so it takes a 48k data file and basically says "use 146 samples out of 147, and then skip that 147th one", at least that's a simple approach that some software might take.  There are more exotic algorithms than that, but basically they are all trying to solve a tough mathematical problem that doesn't need to be a problem in the first place!  http://en.wikipedia.org/wiki/Sample_rate_conversion

Most of the time I record at 24/48 because I really do upload some of those 24/48 flac files to the LMA, and I burn DVD backups of them for "someday".  Rarely, I will even do that at 96k because of peer pressure.  But, the last show I went to, I knew that would not be the case... so I recorded at 44.1k

[digifish_music]

I think he's talking about twice the sampling frequency for the working bit stream, not twice the highest frequency he intends to reproduce.  88.2 is 4X the highest audible frequency one would expect to reproduce from a CD. 

The usual argument for processing at the higher rates and then down sampling is that the digital processing (filters, EQ, etc) produce better results when working with more data.  I don't have any experience with post processing tools so I don't know if any difference in product is audible.

I was indeed just making the sample conversion to 44.1 or 48 kHz (theoretically) more accurate by sticking to an exact multiple of the final frequency (working habit not based on any audible difference). I only record above 44.1 kHz if (as I noted) I am going to do some sample manipulation and slow the recording down, so that ultrasonics become audible, such as you may do when recording Bats or smashing glass for example.

It has always intrigued me when I see how much signal manipulation goes on in the studio (and the miles of cables/patchbays/gear/desks etc) the fuss people make over details of digital audio

I will remain on the record as rarely hearing any meaningful difference between good preamps (except for noise levels) and gear. Particularly in comparison to the huge difference mic placement & post production (can/does) make to the sound. I used to think I did, but got into the habit of blind listening as the final arbiter.

digifish

[boojum]

It seems sampling opinions can be geographic.  On other boards folks will argue, and prove, that hgiher sampling rates do make a difference, and they can hear them!  I still am not sure.   

[digifish_music]

It seems sampling opinions can be geographic.  On other boards folks will argue, and prove, that hgiher sampling rates do make a difference, and they can hear them!  I still am not sure.   

Prove? I have (as yet) never seen any properly conducted experiment that showed people can tell the difference between any sample rate above 44.1 kHz or bit depth above 16 (all else being equal), the data points to the contrary. Got links?

digifish