Up sample Or not

[Jarrett]

I have my own opinions on this one but I would like to hear some other feedback- Situation. Charity work for someone on a pro tools LE system. The only copy of the record is a cdr- Would you upsample to 88.2 (hoping to get better processing power and risk the down conversion on the way out) OR would you leave it 44.1 16 - to avoid the conversion all together? There is no budget really but the record is being put out by a cool label- Typically I am hardcore about the right gear for the job but this is a special situation. If you answer I would also like opinion variations if were were using an HD rig as opposed to the host based processing of an 002 unit. thanks I am looking forward to some opinions-


PS just for the sake of humor I should also mention that the recording engineer(who is in Japan and unreachable) has already crushed it with his Maxim plug-in so truthfully I might do nothing to this except get the sequence in order burn a master and error test it for manufacture-ever happen to any of you

[Bob Boyd]

It depends. A couple of things:

1. upsampling itself doesn't always help the source though I have heard some D/A's open up a bit more above 44.1k.
2. SRC quality can vary greatly.

I will sometimes SRC everything up to 88.2/24 or 96/24 if all the mixes come in at varying rates but I do that just for ease while I'm working on the project.

[Jarrett]

Wow thank you for replying (I was wondering if I had a cyber palgue or something :).


In this case I am importing right off of the disc- Not a typical load in. Of course there is nothing to be gained in the source by up sampling during a direct import in my opinion. - What my specific thoughts are...Do you gain anything in sonic quality of processing by working at a higher samaple rate for instance - Does a waves limiter for example sound better at 88.2?(in an effort to avoid the waves limiter sonic quality discussion lets just say it is whats on hand for this low fi application) That is what I am looking for opinions on- and of course if any quality is to be gained do we lose that in the down sampling process back to 44.1 for manufacture.

The particular project is really great it was just done in someones home in japan- l1 or l2d to death before I got it with noticable distortion- Coming out on a cool label and the music is really interesting.

I -having to obviously do some sort of distortion removal as well as well some other things- am looking for the best quality I can get from the host based lower level pro tools system this must be done on. This one is funny though. it is sitting at the top meter wise and has nothing left for dynamic range at this point( 3 db and I am not really an advocate of the crushing) but it does not have a very full sound - I have a way to fix that but so filling it out a bit will be easy. in referance to all the processing that I need to do...Whats the opinion of my fellow Engineers...Plug in processing gets better in higher sampling environments ...OR no...Or yes but we lose it again on the way down..

[lucey]

Originally Posted by Jarrett

PS just for the sake of humor I should also mention that the recording engineer(who is in Japan and unreachable) has already crushed it with his Maxim plug-in so truthfully I might do nothing to this except get the sequence in order burn a master and error test it for manufacture-ever happen to any of you

I process analog so the sample rate is not an issue. There are many threads on upsampling and SRC if you search around, and that may explain why your replies are few.

If slammed, I might eq-only, and using a tube eq, look to lower the RMS a db or so to fake a sense of more harmonic/dynamic life. And/Or ... use MS comp with a super fast release on the mid and blend the dry mid with the fast released mid.

[taturana]

i think if you are going to process it digitally you could benefit from a good quality src (voxengo r8brain is my choice) for 24/96 upsampling before the processing... especially on eq´s comps and reverbs (strictly IMHO)

lucey, if you process it analog, how´s is your AD setup and what sampling rates?... would you go directly to 16/44.1 or 24/96 and then src for the CD master?

[lucey]

Originally Posted by taturana

lucey, if you process it analog, how´s is your AD setup and what sampling rates?... would you go directly to 16/44.1 or 24/96 and then src for the CD master?

If I'm sure about everything I'll use 16/44.1 and the Rock dithers in the Pacific. If I'm less sure then 24/44.1. This is pretty common practice, unless you're doing digital eq after conversion, then some go 24/88.2 for the eq.

[Bob Boyd]

Originally Posted by Jarrett

Wow thank you for replying (I was wondering if I had a cyber palgue or something :).


In this case I am importing right off of the disc- Not a typical load in. Of course there is nothing to be gained in the source by up sampling during a direct import in my opinion. - What my specific thoughts are...Do you gain anything in sonic quality of processing by working at a higher samaple rate for instance - Does a waves limiter for example sound better at 88.2?(in an effort to avoid the waves limiter sonic quality discussion lets just say it is whats on hand for this low fi application) That is what I am looking for opinions on- and of course if any quality is to be gained do we lose that in the down sampling process back to 44.1 for manufacture.

The particular project is really great it was just done in someones home in japan- l1 or l2d to death before I got it with noticable distortion- Coming out on a cool label and the music is really interesting.

I -having to obviously do some sort of distortion removal as well as well some other things- am looking for the best quality I can get from the host based lower level pro tools system this must be done on. This one is funny though. it is sitting at the top meter wise and has nothing left for dynamic range at this point( 3 db and I am not really an advocate of the crushing) but it does not have a very full sound - I have a way to fix that but so filling it out a bit will be easy. in referance to all the processing that I need to do...Whats the opinion of my fellow Engineers...Plug in processing gets better in higher sampling environments ...OR no...Or yes but we lose it again on the way down..

The only way to answer this is "It depends." Some processes do and some don't. On the higher end of things, the Weiss EQ1-MkII (and EQ1-LP and EQ1-DYN) double sample internally when receiving a 44/48 input. I owned this box before it even went above 48k and the double sampling made a noticable difference. The top end opened up and the low end tighened.

The GML HiRes EQ plugin also double samples internally when processing a 44/48 input. On the other hand, in conversations I had via email with Paul Frindle about his implementation of the GML 8200 in the Oxford EQ, he said that the plug didn't need to upsample because it's calculations were fully resolved at the native rate.

Some mastering engineers think that digital limiters sounds better at higher sample rates and often they do. The potential problem is that if you then follow your limiter with an SRC to convert down to 44.1, you can often get intersample overshoot.

Again, unless you have a really great SRC, I would let that be the determining factor. If you have mediocre conversion, simply avoid it. The bad can outweigh the good pretty quickly.

At the end of the day, it gets back to the way many of these threads end. You'll have to experiment and see what works best to your ears and your workflow with the tools you have on hand.

[Jarrett]

Thank you for the input and opinions. I think expierimentation would be king I have a feeling that your right it is software dependant some things improve some not.

I am actually asking for new prints with no limiter on it. Its just to bad to deal with -(its not as good as some but I did about 4 levels of various types of clip/ crackle removale apps on the top and I couldn't get rid of the distoriton at all
...either that or not put my name on it.


Such a lame practice the Stereo buss destruction before mastering- I already read extensive discussion on this one here yesterday...I think the SSl quad across a mix is a blessing next to over done smashed L2ism. (of course none would be nice too) Not a bad dither I guess but it makes some nasty noise up top in the wrong hands-Straight up dangerous. Or maybe it was Maxim- Iam sure it was one of those wonderful instant masters the consumers are so fond of. LOL

[Otis Grapsas]

Nonlinear processing (analog emualtion, tube emulation, dynamics) is improved by higer sampling rates because the bandwidth is increased, so that more distortion can fit without aliasing because it can be safely be filtered out before reducing to the final rate.

Linear processing benefits when we are talking about filters that are affected by nyquist. By this I mean filters that distort their shapes near nyquist. Bells become assymetric, lowpass becomes higher q etc. At higher rates this is less of a problem. But the calculation noise of the process is increased because the delay lines are longer in samples although they preserve their duration. With some implementations it could be an issue. Technically higher rate means higer calculation noise if the filter structure is kept the same.

There is always a cost form the added processing of sample conversion. If you oversample and downsample just once, the penalty is small. Any processing device that oversamples has this loss. Stacking multiple oversampled processes reduces audio quality. I'm talking about the plugins/outboard that claim to process at a higer rate internally. Put many devices of this type in the chain and you get a hit in audio quality. Is the quality hit important? It depends on the project.

There are techniques to create filters that do not have problems near nyquist while avoiding the quality and performance hit of oversampling. The solution for common bell filters is in the public domain and is used by many plugin manufacturers. By those who are aware of the solution at least:)

[Otis Grapsas]

The oxford GML solution is single rate and this is visible in the frequency graph. If you flip the high shelf horizontally and compare with a low shelf they don't match if you are at 44.1khz. There is visible shape distortion from the effect of approaching nyquist.

[Jarrett]

Thanx Otis That gives me something to think about for a minute. Actually that opened up alot of thought on the subject. Thank you a whole lot for the input.

[Otis Grapsas]

I just noticed something else in your post. A conversion never has an advantage by itself, even if you keep the material at the high rate. The same applies for conversion of bit depth. The material actually loses quality when getting from 16/44.1 to 64bit float/192khz. We only convert if the processing that will follow will benefit from the higher data format.

Users usually think oversampling is a good thing, but I spend lots of time and effort to ellinimate the need for oversampling/downsampling and I know other developers that strive for the same. From a design point of view, oversampling is a brute force method. It saves you the research and development time of improving the process. You can use common coefficient generation techniques for an EQ instead of developing something more advanced or make a mediocre dynamics tool and a tube emulation look better in measurements. You can make it measure better without even touching its nonlinear elements or changing the sound in any way (except alias)! The truth is that it doesn't help much. Yes, there will be less aliasing, but the harmonics produced will still sound bad. By the time alias becomes audible, the process is already too deep into harshness to let the oversampling make any difference. If your bass has tons of bad sounding distortion at 300hz to 1khz, you don't really care about a little alias at 10khz that is usually masked by the program. You will actually prefer a process that aliases there but produces good sounding distortion on bass or the midrange. The better process will still benefit from oversampling of course but this is something that has to be balanced against the loss of oversampling/downsampling cycle. A tube emulation can use oversampling to push nyquist higher, but even if you oversample 1000x the actual sound of the process will not change. If it does need high oversampling in order to not sound terrible, you can be sure the process is bad. A better tube emulation will produce low orders of distortion that decay fast from one order to the next, especially as you get lower in input amplitude. This process will not alias much and it will sound a lot better without oversampling.

[lucey]

Originally Posted by Otis Grapsas

Users usually think oversampling is a good thing

aargh, marketing