R/E/P Community

Please login or register.

Login with username, password and session length
Advanced search  

Pages: 1 [2] 3  All   Go Down

Author Topic: 24/96 to 16/44.1 - Best Method?  (Read 5708 times)

bbalin

  • Newbie
  • *
  • Offline Offline
  • Posts: 8
Re: 24/96 to 16/44.1 - Best Method?
« Reply #15 on: June 06, 2004, 02:45:45 pm »

Sorry for jumping in here - but this topic really makes me curious :
@Brad :
Quote:

At this point, I avoid SRC like the plague as I have found nothing in software that is invisible

Brad, what are the problems you're seeing with software SRCs ?

I remember our (small) discussion on the samplitude forum but unfortunately it did not go on.

-Boris
Logged

bblackwood

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 7036
Re: 24/96 to 16/44.1 - Best Method?
« Reply #16 on: June 06, 2004, 04:06:49 pm »

bbalin wrote on Sun, 06 June 2004 13:45

Brad, what are the problems you're seeing with software SRCs ?

I remember our (small) discussion on the samplitude forum but unfortunately it did not go on.

Hi Boris, welcome to the board...

When using any software SRC I've used, I notice a slight collapse of the soundfield along with an overall darkening of the track. The very highest setting in Samp7, for example, yields a narrower soundfield an sounds like you reduced 10kHz by a good .5dB. Not acceptable to me.

The best hardware SRCs do not do this, but even the finest ones (Weiss and Lavry) use chips that have been surpassed.

Once I figure out whether the new box works or not, I'll let you know...
Logged
Brad Blackwood
euphonic masters

OatBran

  • Newbie
  • *
  • Offline Offline
  • Posts: 21
Re: 24/96 to 16/44.1 - Best Method?
« Reply #17 on: June 06, 2004, 11:25:22 pm »

  I have a quick question that applies indirectly to this topic.  In my experience I have always found the conversions from 88.2 to 44.1 more seemless then the 96 to 44.1 conversion.  Not much more, but apparent none the less.  I haven't used any hardware src to compare with, but all of my listening with the software side on various projects has shown the few 88.2's I have gotten to be a little cleaner transition then I was able to get with the 96's.  Most of these projects with higher sampling rates have been orchestral in nature... maybe that has something to do with it.  I am guessing that the reasons for this might have something to do with a more linear conversion (x2).

 Is this just in my head/ears or has anybody else found this?  And if so, why did we as an industry go 96 instead of 88.2 if our market is still 44.1?
Logged

TotalSonic

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 3728
Re: 24/96 to 16/44.1 - Best Method?
« Reply #18 on: June 07, 2004, 12:21:08 am »

OatBran wrote on Mon, 07 June 2004 04:25

 
 Is this just in my head/ears or has anybody else found this?  And if so, why did we as an industry go 96 instead of 88.2 if our market is still 44.1?


I think the 96kHz decision might reflect more the fact that digital film/video production has been using 48kHz as its standard for quite some time now.  So I guess the market envisioned is more towards video standards than audio ones.  Kind of reflects the trend that DVD-V sales are going up while CD sales are pretty flat. Cue The Buggles "Video Killed The Radio Star"

Best regards,
Steve Berson

OatBran

  • Newbie
  • *
  • Offline Offline
  • Posts: 21
Re: 24/96 to 16/44.1 - Best Method?
« Reply #19 on: June 07, 2004, 05:29:47 am »

  Ah, indeed!  A very good reason and I now remember having that presented to me as an answer before...

Thanks Very Happy
Logged

bbalin

  • Newbie
  • *
  • Offline Offline
  • Posts: 8
Re: 24/96 to 16/44.1 - Best Method?
« Reply #20 on: June 07, 2004, 04:53:06 pm »

Quote:

When using any software SRC I've used, I notice a slight collapse of the soundfield along with an overall darkening of the track. The very highest setting in Samp7, for example, yields a narrower soundfield an sounds like you reduced 10kHz by a good .5dB. Not acceptable to me.


I have yet to do these tests myself.
As I do not own 88/96k capable equipment up to date I was limited to bench testing using upsampled or synthetic signals.
What I found out :
When upsampling : Both Wavelab and Samp do create alias tones and do not have a proper passband filter, Wavelab being worse than Samp (in highest quality setting).
On the contrary Cool Edit Pro (now Adobe Audition) performs nearly ideal in this respect.
Incomplete filtering also is a problem when downsampling in Wavelab and Samp, again Cool Edit Pro is yielding smooth results.

As for clipping/overload after SRCing I'm pretty sure that you simply cannot avoid it (depending on the type of source material of course) because the SRC has to deal with the intersample peaks and the shifted sampling points will hit the waveform at some arbitrary point.
As you know using 32bit float format will overcome at least the clipping.
In this respect Cool Edit Pro offers a surprise again : it has a waveform display actually showing the intersample peaks so you can easily watch the potential problems before conversion.
And you could modify the settings of your limiter/maximizer Wink

Boris
Logged

mark4man

  • Full Member
  • ***
  • Offline Offline
  • Posts: 112
Re: 24/96 to 16/44.1 - Best Method?
« Reply #21 on: June 07, 2004, 08:09:13 pm »

Quote:

In my experience I have always found the conversions from 88.2 to 44.1 more seemless then the 96 to 44.1 conversion. Not much more, but apparent none the less. Is this just in my head/ears or has anybody else found this?

Yes.  88.2 to 44.1 is a direct (even) divisible (not the correct mathematical term...but all I can think of, right now.)  The SRC is optimized for integer scaling with direct multiples.  Far less rounding errors.

Oh..it's Integer Factor.  88.2 to 44.1 is represented by an integer factor of 2:1.

mark4man

BTW - I know my post was relative to 96 to 44.1...but that's because everything our band members give me to produce is 24.96.  They do not have 88.2 capability.
Logged

OatBran

  • Newbie
  • *
  • Offline Offline
  • Posts: 21
Re: 24/96 to 16/44.1 - Best Method?
« Reply #22 on: June 07, 2004, 09:06:17 pm »

Thanks mark4man.  That was my understanding as well.  So why is it then that the music industry is still using 96k?   I understand the video aspect and its compatibility, but rarely is that an issue with most/any of the projects I deal with.  Not sure about you guys.  But if 88.2 sounds just as good as good as 96 (or damn close), then why would we use 96k and allow this src to be so nasty?  Should we be using 88.2 unless we know that video might be involved?  Would more mastering engineers prefer that?  Do 9 out of 10 dentists agree?

Thanks.
Logged

bblackwood

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 7036
Re: 24/96 to 16/44.1 - Best Method?
« Reply #23 on: June 08, 2004, 02:22:28 am »

Well, iirc, most modern async SRCs upsample to a very high rate then convert to the lower rate. The benefit of 2:1 ratios only exists in sync converters.

Again, feel free to correct, but I'm pretty the good asyncs can do it equally well...
Logged
Brad Blackwood
euphonic masters

Doug Van Sloun

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 62
Re: 24/96 to 16/44.1 - Best Method?
« Reply #24 on: June 10, 2004, 10:01:53 pm »

OatBran wrote on Sun, 06 June 2004 22:25

  I have a quick question that applies indirectly to this topic.  In my experience I have always found the conversions from 88.2 to 44.1 more seemless then the 96 to 44.1 conversion.


I think that this points to the algorythms at work and the precision with which they are carried out.  With the Weiss SFC2 I can detect no difference going between any of the 4 sample rates offered, up or down.
Logged
Doug Van Sloun
http://www.focusmastering.com
Omaha, NE

chrisj

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 959
Re: 24/96 to 16/44.1 - Best Method?
« Reply #25 on: June 11, 2004, 03:18:38 pm »

I'm honestly surprised by this thread as well- Brad's way ahead of me in so many things that I'm puzzled. Maybe I can help. Try this:
http://www.mega-nerd.com/SRC/
This algorithm is present in the open source tool 'Audacity'. It's based on an algorithm by Julius O. Smith using (insert geek block HERE) windowed sinc interpolation at an arbitrary quality level- you trade off between speed and quality.
What this means is, I've fooled with a lot of poorer SRC and done things like tested them with a sine sweep from 10K to 26K (at 48K sampling rate) and converted it to 44.1K. You get really LOUD birdies and artifacts if you do that. Normal software SRC is appalling under those circumstances, or if you have particularly nasty abrasive program material as I've sometimes had. The windowed sinc SRC I mentioned above successfully converts the sweep-to-above-20K so well that you still (I still...) can't hear the sine as it goes above about 15K. Any other software SRC (maybe not Barbabatch) and you totally hear lots of loud artifacts the whole time.
The quality settings work so that at low quality you still don't get artifacts, but it starts rolling off at 14-16K. At the very highest quality you can get arbitrarily close to perfect passband up to the Nyquist frequency, and still perfect rejection after that- absolutely no audible artifacts.
The concept is called an emulated conversion of the original data to analogue and then resampling at the new sample rate- which is of course what you're doing now. If this sound library is helpful to you, please let me know! I'd like to think I was able to come up with something to help you out, and I know this algorithm performs fantastically. I'd love to see other software incorporating it (it would have to be open source to incorporate this 'Secret Rabbit Code' library, but again there is already usable OSS software that does, like Audacity)
Good luck! I suspect it should be possible to do this operation in the digital domain as well or better than in analog, because you're not supposed to be adding any color- it's all about rejecting the artifacts, and it's all about finding ways to throw more computational power at the problem. Windowed sinc has that capability, and should solve your problem.

chrisj

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 959
Re: 24/96 to 16/44.1 - Best Method?
« Reply #26 on: June 11, 2004, 03:30:10 pm »

Oh- an interesting side effect of the windowed sinc is that you can generate a sweep at 48K and have it stay fairly audible throughout, due to artifacts that apparently interact with the D/A in some way. Generators aren't always perfect. You see the meters on the power amp light up (if you got 'em) and hear a high pitched sweep that's actually supported by artifacts. Convert with windowed sinc to 44.1K, and the sweep becomes inaudible, meters still light, and if you look at the waveform it's plainly still a sweep, but you've cleaned up artifacts present in the original file. I'm not entirely sure how that works, and doubt it would happen if you weren't downsampling, but it does...

bblackwood

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 7036
Re: 24/96 to 16/44.1 - Best Method?
« Reply #27 on: June 11, 2004, 03:42:24 pm »

chrisj wrote on Fri, 11 June 2004 14:18

I'm honestly surprised by this thread as well- Brad's way ahead of me in so many things that I'm puzzled. Maybe I can help.

Umm, I don't get it - what does this mean?

FWIW, I'm quite sure I'll end up with a hardware async SRC when all is said and done...
Logged
Brad Blackwood
euphonic masters

chrisj

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 959
Re: 24/96 to 16/44.1 - Best Method?
« Reply #28 on: June 11, 2004, 04:10:09 pm »

bblackwood wrote on Fri, 11 June 2004 15:42

chrisj wrote on Fri, 11 June 2004 14:18

I'm honestly surprised by this thread as well- Brad's way ahead of me in so many things that I'm puzzled. Maybe I can help.

Umm, I don't get it - what does this mean?

FWIW, I'm quite sure I'll end up with a hardware async SRC when all is said and done...



Easy- it means 'hey, that's weird- I'm using this SRC that's essentially perfect and Brad doesn't seem to have heard of it'. Maybe it's because I'm hanging around with the OSS programmers where the idea germinated? Some of you guys are treated with such, um, reverence (excepting Mixerman of course) that it's surprising to see you hitting a wall and not having solutions to a thing. At the same time, I totally understand your finding software SRC inadequate, because I've done the tests and even experimented with SRCing through extraordinarily high intermediate sample rates, manually. It doesn't really help, and the windowed sinc does.

If you can get a hardware async SRC, it would be because that is using some type of similar algorithm (unless it's literally got an analog stage, which would be an interesting approach and I'd be fascinated to hear about anybody doing that- an analog stage that was about a half inch of circuit board traces heavily shielded, that would be a heck of a good way to bounce off the analog domain).

Since libsamplerate ('Secret Rabbit Code', as used in Audacity) is _software_ async SRC using essentially a virtual analog intermediate stage and NOT just a very-high-rate intermediate stage, I still think you should check it out. Honestly, it performs very well. Very very well. I am dubious a hardware box could be better sonically, though I'm sure one could be as good.

bblackwood

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 7036
Re: 24/96 to 16/44.1 - Best Method?
« Reply #29 on: June 11, 2004, 04:31:40 pm »

chrisj wrote on Fri, 11 June 2004 15:10

Since libsamplerate ('Secret Rabbit Code', as used in Audacity) is _software_ async SRC using essentially a virtual analog intermediate stage and NOT just a very-high-rate intermediate stage, I still think you should check it out.

What does that mean? The best SRCs I've seen do simply upsample to a very hi fs before downsampling. I might have missed it but what on earth is 'a virtual analog intermediate stage'?
Logged
Brad Blackwood
euphonic masters
Pages: 1 [2] 3  All   Go Up