Topic: Re-sampling (Read 4246 times)

Oldfart · « **on:** July 31, 2004, 10:31:48 PM »

Hello everyone,

a wise man once told me, there are no stupid questions, just bad answers.

So here's my question of the month: when is it best time to re-sample?

As usual , thank you in advance (if you reply),

Oldfart

craig boychuk · « **Reply #1 on:** August 01, 2004, 01:24:04 PM »

To add to the question:

Maybe this should be rudimentary knowledge, but I don't think I've ever gotten a straight answer about this...

Do you gain anything from recording at a high sample rate when the destination format is CD?

I suppose if you are multitracking at a high sample rate, but mixing to an analogue format it makes sense. However, when you have to convert down to 44.1, is there really any difference than if you were tracking at 44.1 in the first place?

TotalSonic · « **Reply #2 on:** August 01, 2004, 01:36:41 PM »

I've yet to hear any software based src that wasn't degrading at least a little bit. I usually prefer to do a loopback da to ad for sample rate conversions to avoid the math aspect - although I don't have the Weiss or Lavry boxes so I can't make judgement as to whether these do superior to loopback. In my experiences any "gains" from recording at 48kHz if the destination is 44.1 are not audible in the final product. For well made 96kHz recordings I think you're going to be able to still hear some of the benefits of having a higher res source by doing all processing at 96 and then applying the src and dithering down as the final step.

Best regards,
Steve Berson

Innominandum · « **Reply #3 on:** August 01, 2004, 02:22:56 PM »

In good software SRC, the only degrading conversion is downsampling. In good software SRC, upsampling has absolutely no (0) effect on sound quality.

The only thing that causes degradation in [well done] software downsampling is the anti-aliasing filter. The filter ends up chopping off a bit of the high end. In my case it was 20.5 kHz to nyquist.

I subtracted a downsampled version from a perfect target version. The only difference between the downsampled file and the target file was in the aforementioned frequency range. I played the resulting difference file full blast and couldn't hear anything.

If I had the choice between good hardware SRC and good software SRC, I'd pick the hardware. But if you can't afford one of those expensive units, I wouldn't sweat it.

bobkatz · « **Reply #4 on:** August 01, 2004, 04:22:22 PM »

Everything's a compromise! As good as the Weiss SRC is, it's not 100% transparent. But sometime between 5 and 10 years ago things changed for me in favor of upsampling in the mastering chain (either via digital or analog) and/or recommending originating at the higher rate. It used to be that "start and stay at 44.1 kHz because the SRCs are so bad" was very true. But now for me it's "upsample to a higher rate and work at that rate for as long as you can before going back to 44".

I also feel that projects that come to me at 48K and above sound marginally better (on the average) than those that come in at 44.

Basically, the old principle remains of "the higher the quality of the source, the higher the quality of the final result". It's just that for a long time we couldn't realize that principle because the losses of resampling were so high. But not anymore. DCS, Weiss, and Lavry make excellent SRCs. The new Z Sys SRC is miles above the previous one, by the way, and is not very far below the Weiss, in my opinion. We're talking about a chip change from the AD 1890 to the AD 1896.

David Glasser · « **Reply #5 on:** August 01, 2004, 04:33:26 PM »

I agree with all of Bob's points, and conclusions. In addition to the units mentioned, the Prismsound ADA-8 also has excellent synchronous multichannel src - something that previously required multiple dcs boxes. I'll also add that the background SRC built into SonicStudio HD is about the best I've heard - in software or hardware, and have built my typical workflow around using that (if src is required) after all the EQ has been done.

Bobro · « **Reply #6 on:** August 02, 2004, 07:09:34 AM »

Samplitude's built-in SRC, at the highest setting, is remarkable. Takes its sweet time about it though- I go out for a drink or three while it is running.

-Bobro

bblackwood · « **Reply #7 on:** August 02, 2004, 09:35:47 AM »

Bobro wrote on Mon, 02 August 2004 06:09

Samplitude's built-in SRC, at the highest setting, is remarkable. Takes its sweet time about it though- I go out for a drink or three while it is running.

I have had the opposite results, finding that even at the highest setting, the degradation was unacceptable in Samp. Now I'm looking at hardware units.

Guess that just reinforces what they say about opinions, eh?

craig boychuk · « **Reply #8 on:** August 02, 2004, 10:43:39 AM »

What is the fundamental difference between software src and hardware src? It's all math, right? Or does analogue conversion come into play...?

Thank you all for the feedback!

TotalSonic · « **Reply #9 on:** August 02, 2004, 11:40:30 AM »

mid-fi wrote on Mon, 02 August 2004 15:43

What is the fundamental difference between software src and hardware src? It's all math, right? Or does analogue conversion come into play...?

Good question - depends on the box. Some might have internal da/ad paths - but I believe most are using a ton of dedicated high bit rate dsp (the Weiss SFC2 for example uses 40bit floating point / 32bit fixed point processing) so that a number of calculations can be done in real time which would cripple most cpu's. As native processing power has markedly increased in recent years (i.e. with a recent update my SAWStudio workstation can now do a "high quality" src for a single stereo track in real time using double precision 32bit fixed point math on a 2.8GHz Pentium) the real difference then becomes the algorithm used.

Maybe users of other src boxes like the Lavry can chime in here with the basics of their processes - I'm very curious myself on this.

Best regards,
Steve Berson

OTR-jkl · « **Reply #10 on:** August 02, 2004, 01:58:10 PM »

bblackwood wrote on Mon, 02 August 2004 08:35

Bobro wrote on Mon, 02 August 2004 06:09

Samplitude's built-in SRC, at the highest setting, is remarkable. Takes its sweet time about it though- I go out for a drink or three while it is running.

I have had the opposite results, finding that even at the highest setting, the degradation was unacceptable in Samp. Now I'm looking at hardware units.

Guess that just reinforces what they say about opinions, eh?

Ditto to what Brad said.

I just finished a project last Fri that came in at 48/24 and did the SRC in both Samp7 and R8B (both set to their respective highest settings). Neither were right on, but R8B was much closer, IMO.

As for the original question:
That's a good question. I did the SRC as the very last thing - after the files were completely processed and dithered. I haven't tried it any other way but would also like to know if there's a better time...

Oldfart · « **Reply #11 on:** August 02, 2004, 02:31:43 PM »

To OTR:

Strange I was always told to do dithering last, but I'll check out.

Thanks,

Oldfart

Bob Olhsson · « **Reply #12 on:** August 02, 2004, 05:32:10 PM »

R8B dithers so there's no need to if you convert to the final format.

I'm impressed that dither isn't an option, it shouldn't be other than as a special distortion effect.

OTR-jkl · « **Reply #13 on:** August 02, 2004, 05:44:02 PM »

Bob Olhsson wrote on Mon, 02 August 2004 16:32

R8B dithers so there's no need to if you convert to the final format.

Are you sure...? Where did you find out? I couldn't find anything on their website or in the plug itself...

bobkatz · « **Reply #14 on:** August 03, 2004, 05:24:14 PM »

OTR-jkl wrote on Mon, 02 August 2004 13:58

As for the original question:
That's a good question. I did the SRC as the very last thing - after the files were completely processed and dithered. I haven't tried it any other way but would also like to know if there's a better time...

Absolutely there's a better time! SRC should be the penultimate (next to the last) process and dithering to 16 bit the last process. The wordlength grows on the output of the SRC...

OTR-jkl · « **Reply #15 on:** August 03, 2004, 11:05:38 PM »

Quote:

SRC should be the penultimate (next to the last) process and dithering to 16 bit the last process. The wordlength grows on the output of the SRC...

Here's my workflow:
I'm using a s/w limiter so the processed (pre-limited) material gets captured at the current project format (48/24 in this case) and dither is then applied parallel to the limiter (if I use the CraneSong dither) or after limiting if I use Samp Trian. or Powr dither.

Are you saying that it would be best if I captured the pre-limited/pre-dithered material at 44/24 instead of 48/24? That would require a real-time SRC by Samp in my case... All I have at present is a custom-loaded tube (Telefunken) DA/AD that I run stuff across that needs some analog flavor added and it is being fed a dig signal from the soundcard and gets routed back to the soundcard via spdif.

The real-time SRC in Samp is not as hi-quality as the offline SRC and the offline SRC set at its highest setting lacks something to be desired....

dcollins · « **Reply #16 on:** August 05, 2004, 01:59:10 AM »

mid-fi wrote on Mon, 02 August 2004 07:43

What is the fundamental difference between software src and hardware src? It's all math, right? Or does analogue conversion come into play...?

Well, it's all in the implementation. Math is in there, though.

There are NO units that do A/D/A conversion, btw.

DC

Level · « **Reply #17 on:** August 05, 2004, 02:03:05 AM »

Quote:

There are NO units that do A/D/A conversion, btw.

Explain?

I have a Sony that takes the direct digital transfer strait to analog without the need to do hard storage and reconvert. It is called a buffer.

dcollins · « **Reply #18 on:** August 05, 2004, 02:12:19 AM »

Level wrote on Wed, 04 August 2004 23:03

Explain?

I have a Sony that takes the direct digital transfer strait to analog without the need to do hard storage and reconvert. It is called a buffer.

There are zero Sampling Frequency Converters that use analog conversion. It is always a completely digital process.

DC

bobkatz · « **Reply #19 on:** August 05, 2004, 11:29:54 AM »

OTR-jkl wrote on Tue, 03 August 2004 23:05

Quote:

SRC should be the penultimate (next to the last) process and dithering to 16 bit the last process. The wordlength grows on the output of the SRC...

Here's my workflow:
I'm using a s/w limiter so the processed (pre-limited) material gets captured at the current project format (48/24 in this case) and dither is then applied parallel to the limiter (if I use the CraneSong dither) or after limiting if I use Samp Trian. or Powr dither.

I understand what you're doing, I think. But then I got confused when you wrote:

Quote:

Are you saying that it would be best if I captured the pre-limited/pre-dithered material at 44/24 instead of 48/24?

I've not completely got your workflow and patching down, but regardless of the processes that you use, your NEXT TO THE LAST STEP should be SRC down to 44/24 (or 44/32) and your LAST STEP is to dither to 16 bit.

There are some fine points if you have to work out of the box at any point (like if you are feeding an EXTERNAL SRC from your 32 bit native program), but let's first see if you can do the above using the equipment you have. If the Samplitude SRC is not very good, then yes, you will have to get a better software or hardware SRC. Hope that's not what's holding you back Sad

BK

OTR-jkl · « **Reply #20 on:** August 05, 2004, 04:09:53 PM »

Bob -
See if you can follow this:

Current chain:
- 1st instance of Samp used for playback of original file and for any processing done In-the-Box (EQ, Comp, etc.) <Playback is at project sample freq/bit depth>
- Dig out of soundcard into tube unit (to add analog flavor)
- Dig out of tube unit back into soundcard
- 2nd instance of Samp used to capture incoming processed audio at project sample freq/32-float (pre-limiting/pre-dither)
- Limiting and dither are applied in the 2nd instance of Samp

Here's the key:
Samp has a Live Input function that allows me to capture an incoming audio signal on the HD while simultaneously applying FX on the incoming channel. Because of this, I can apply the s/w limiter on the channel that is armed to record (or apply the limiter across a buss that the channel is routed to...) and the signal gets processed (limited) on the way to the HD. This set-up allows me to hear all processing working together (just as if it were all being done on outboard gear and I were monitoring the output of the "capture" DAW.

When I get everything set the way I want it, I turn Samp's real-time Mixdown fundtion On and play the source file. Source file gets processed thru the chain and captured on the HD as a finished file.

In using this method though, I still have to deal with the SRC (when applicable). With my current gear, there's only 3 ways I can do it:
- On the way into the "capture" (2nd) instance of Samp which is not a very good real-time SRC
- At the end of the process (post processed/dithered/captured) off the HD using offline Samp SRC or other like R8Brain.
- Pre any processing (first thing - and probably the worst scenario...?)

There is one other possibility:
I could send the signal Dig out of the playback instance of Samp into another piece of gear (like my Tascam CD-RW2000) and use it to do the D/A, then send it to the analog input of the tube unit , then dig out of the tube unit and on to the soundcard dig input into the 2nd instance of Samp capturing at 44/32. I think it would work but I'm leery of using the D/A convertors in the CD-RW.

Normally I don't have to worry about all this as the source files are mostly 44/24. But, as was the case last week, I need to come up with the best way to do the SRC (until I can get an outboard unit).

I hope I explained this clearly enough to be understood...

OTR-jkl · « **Reply #21 on:** August 09, 2004, 05:22:58 PM »

Well, I played around with the SRC on this particular project some more and here's what I found:

Initially, I was capturing the 48/24 processed material at the original sample freq/bit depth then re-sampling down to 44.1 using r8brain at its highest settings.

On the second run, since I was using the Crane Song dither from the Analog Dither CD, I turned the dither off (muted that track) and captured the material at 48/32float. I then placed the 48/32 files on the EDL of a 48k Samp project and placed the CS dither track on the EDL underneath the music track. I did the PQ editing and (making sure that Samp's Re-sampling quality was set to its highest setting) burned a CD.

Result:
Waaaaaay more like the originally captured 48/32 file. Still not exactly the same (as expected), but much closer.

Conclusion:
Re-sampling after dithering makes the top end sort of cloudy or "smeared" sounding which results in a loss of clarity and punch. So, like Bob K said, apply dither as the very last step of the process (but all of ya'll probably already knew that!!).

BTW - Thanks Bob...

Bobro · « **Reply #22 on:** August 15, 2004, 05:17:40 AM »

bblackwood wrote on Mon, 02 August 2004 14:35

Bobro wrote on Mon, 02 August 2004 06:09

Samplitude's built-in SRC, at the highest setting, is remarkable. Takes its sweet time about it though- I go out for a drink or three while it is running.

I have had the opposite results, finding that even at the highest setting, the degradation was unacceptable in Samp. Now I'm looking at hardware units.

Guess that just reinforces what they say about opinions, eh?

Most likely what I'm listening for is too specific and out of date. I live in mortal fear of the blue-silver/Eighties-on-steroids sound that I associate with so many recordings and tools of the last ten-fifteen years, so if I'm getting say synaestasia autumn colors before and after the SRC, that's good AFAIC.

I say out of date because that particular sound I hate doesn't seem to dominate the radio pop I hear in bars any more, nor does it seem to be the "default" signature of new software tools.

What do you specifically find to be degraded?

-Bobro

R/E/P Community

Author Topic: Re-sampling (Read 4246 times)