Topic: 24/96 to 16/44.1 Best Practices & State of the Art Software?

[Chilly Brioschi]

Hey all you DAW gurus!

I haven't changed my workflow or software in a few years.  (CEP/Audition 1.5 with a 0.5 bit noise dither)
Knowing the way technology moves, I'd guess that I'm missing something.

What is the best way to convert 24/96 to 16/44.1 these days?

Which Freeware does it best?

All stops out, which commercial software does it best?


TIA

[stevetoney]

You might want to define what you mean by 'which does it best'.  I assume that you mean more accurately. 

I'm sure I'm not qualified to answer this question, and I _KNOW_ I'm no guru, but I do know I haven't read anything that would compell me to think that conversion is no better OR worse in Audition 3.0 than in Audition 1.0 in terms of accuracy. 

Having said that, Audition 3.0 seems to have better workflow than 1.0 in that most processes are performed faster on my machine.  So if the definition of better is based on speed, then 3.0 is better than 1.0.  3.0 had some enhancements in terms of basic workflow features (fewer keystroke types of features), but honestly I didn't feel that there was all that much core feature-wise in 3.0 that isn't also in 1.0, but I'm not a power user for sure.

Having further said the above, I find that Audition 3.0 is a tradeoff on my machine because it seems to be a real RAM hog.  Slows down multi-tasking and causes frequent lockups on my machine. 

In the end, because of all of the above combinations of considerations, I've personally opted to stick with Audition 1.0.  If I had a bigger, newer machine, I'd probably go with 3.0.

Have no idea about other software packages, so cannot speak to those.

[live2496]

You might want to take a look at these graphs...
http://src.infinitewave.ca/

As far as free programs R8Brain (free) is an excellent program for resampling.
http://www.voxengo.com/product/r8brain/

Gordon

[stevetoney]

You might want to take a look at these graphs...
http://src.infinitewave.ca/

I know this isn't my thread, but I'm trying to learn too.  I'm not sure what the graphs should be telling us.  Can you please explain? 

First question is what is SRC?

Regardless of that though, I will try to at least state a little bit what I see.  It's obvious looking at the top graph vs. the bottom one that there are more lines in the top one (do stray lines = sample noise???)...the top says it's unfiltered while the bottom says its a filtered sample, so it looks like a filtered conversion from 96 to 44.1 yields maybe less noise???  (makes sense...filtering implies taking something like noise out of the sample)

But then, both tests are Audition 2.0.  So, since the original poster was asking about upgrades to software, are you also saying that Audition 2.0 includes this filtering feature and Audition 1.0 does not, or does it better than Audition 1.0? 

I guess I'd just like to have someone explain please what this information should be telling us...and more generally are these graphs or is this website used to determine how effective the various tools we use are at accomplishing the various functions that they perform?  If so, is there someplace that explains to dummies how to use the website.

[Lil Kim Jong-Il]

As far as free programs R8Brain (free) is an excellent program for resampling.
Gordon

I have a question about r8brain that was not addressed on their forum: 

Is there a way to perform batch conversions so that the sliding window will gracefully span file boundaries?

I'm thinking of the case in which a 24/96 recording is already split into tracks and the tracks are to be down converted to 16/44. 

[Gutbucket]

You might want to take a look at these graphs...
http://src.infinitewave.ca/

I know this isn't my thread, but I'm trying to learn too.  I'm not sure what the graphs should be telling us.  Can you please explain? 

First question is what is SRC?

SRC = Sample Rate Conversion 

The link above is to a comparison of various conversion algorithms.
There is a help button on the page that explains each test and graph.

[stevetoney]

Thanks gutbucket.  That helps!

[live2496]

Regardless of that though, I will try to at least state a little bit what I see.  It's obvious looking at the top graph vs. the bottom one that there are more lines in the top one (do stray lines = sample noise???)...the top says it's unfiltered while the bottom says its a filtered sample, so it looks like a filtered conversion from 96 to 44.1 yields maybe less noise???  (makes sense...filtering implies taking something like noise out of the sample)

But then, both tests are Audition 2.0.  So, since the original poster was asking about upgrades to software, are you also saying that Audition 2.0 includes this filtering feature and Audition 1.0 does not, or does it better than Audition 1.0? 

I find the frequency sweeps the most interesting. If you see what I would call a ghost image this indicates aliasing artifacts. Sample rate conversion down from a higher sampling rate to a lower one requires bandwidth filtering. A good filter (and a good algorithm I presume) would give you a clean sweep with no artifacts.

I wasn't making any inferences about Adobe Audition... I just thought I would present the link to the information and let the reader delve into it and make their own conclusions.

[live2496]

Is there a way to perform batch conversions so that the sliding window will gracefully span file boundaries?

Not sure if I follow what you mean by sliding window? Is this in the pro version?

I have just used the free version and found it to be more than adequate for my needs. In the free version there is a batch conversion button that will convert all files in a folder. (I hope I am understanding your question )

P.S. The free version also includes a DLL. The developer has stated that if anyone wanted to do a command line tool for this they could do so by calling functions in the DLL. A command line based conversion tool should be possible if someone wanted to make one. That way scripts could be written to batch convert files.

[Gutbucket]

What I'd like to get a better handle on is the trade-offs made in various SRC designs.  I'm not sure exactly how a nice clean looking graph translates into sound quality and I really have no idea what the sonic implications are of improving one aspect (like say the aiming for a nice flat phase response) while sacrificing another (like introducing pre-ringing in the impulse response).  I really wish that site had sound samples that we could listen to in order to really understand what's going on and what those design trade offs mean.

[Lil Kim Jong-Il]

Is there a way to perform batch conversions so that the sliding window will gracefully span file boundaries?

Not sure if I follow what you mean by sliding window?

Lots of noise shaping/dithering algorithms do not work on discrete samples but work on a group of neighboring samples to calculate the value for a specific sample result.  As the file is processed, this group moves along the file.  That moving group of data is called a sliding window. 

The issue I'm considering is that there is always some non-ideal performance of the algorithm when the window is not full.  So at the beginning or end of a stream when the window is being filled or being drained, the results that are produced are not the same results that would be produced if the window was constant.  This isn't an issue if the signal fades in and then fades out because the window can be padded with 0s and the behavior is correct for a signal with a slightly longer leader or trailer.

Where a problem occurs is when a show gets tracked and each track is batch processed as a discrete stream sequence.  The sliding window sees a discontinuity - most likely a step impulse - at the beginning and ending of each track.  When the batch processed tracks are concatenated at playback, there is discontinuity in the noise shaping at each track boundary. 

I don't know if it would be audible but it is there and I consider it less than ideal.  Samplitude addresses this by creating a virtual contiguous stream for processing of multiple files concatenated and track bounced.  Any tool could be written to do this since it's just a matter of saving some intermediate results when the window slides across the file boundaries.  I don't know about r8brain, so I was asking.  There are times when my workflow would be easier if I could just cut everything up with CDwave and batch process the 24/96 files to 16/44 instead of using samplitude.  Or when I get 24-bit master tracks from someone and want to make CDs.

[Lil Kim Jong-Il]

I really wish that site had sound samples that we could listen to in order to really understand what's going on and what those design trade offs mean.

Didn't Brian do this and post the results here somwhere?  If he still has the master track anyone could extend the comparison set as new tools are released.

edit: here it is http://taperssection.com/index.php/topic,51478.0.html

[live2496]

What I'd like to get a better handle on is the trade-offs made in various SRC designs.  I'm not sure exactly how a nice clean looking graph translates into sound quality and I really have no idea what the sonic implications are of improving one aspect (like say the aiming for a nice flat phase response) while sacrificing another (like introducing pre-ringing in the impulse response).  I really wish that site had sound samples that we could listen to in order to really understand what's going on and what those design trade offs mean.

If you follow the discussions on Bruno Putzey's forum on REP they may discuss stuff like this from time to time.  
example: http://recforums.prosoundweb.com/index.php/t/19917/12697/

Also dither and sound shaping. Most of it way over my head.

[Gutbucket]

Thanks live & Kim Jong, I'll check those out.  Hopefully I won't get overly deep.

'Kim, do you personally use the SRC in Samplitude or output files to R8Brain? and do you hear a difference?  I'm just getting into Sam and have wondered about which way I'll end up going work-flow wise. I'll comp it at some point, just wondering about your opinion.  Sure would be nice if I don't hear a difference. 

[Lil Kim Jong-Il]

'Kim, do you personally use the SRC in Samplitude or output files to R8Brain?

I use samplitude all the way through.  I have a workflow that starts with the raw 24/96 from the 722 and ends with 24/96 and 16/44 track sets.  Then I flac them.  I wish I knew of a flac plugin for samplitude.


I have not done a lot of comparisons because I generally listen to 24-bit at home and MP3 in the car.

[Gutbucket]

I've been wishing the same about FLAC support in Sam.  Haven't gotten around to exploring SRC further because I too mostly just play my 24/48s off the HD or SD cards.   Just a bit low on the priority list.

[live2496]

The issue I'm considering is that there is always some non-ideal performance of the algorithm when the window is not full.  So at the beginning or end of a stream when the window is being filled or being drained, the results that are produced are not the same results that would be produced if the window was constant.  This isn't an issue if the signal fades in and then fades out because the window can be padded with 0s and the behavior is correct for a signal with a slightly longer leader or trailer.

Where a problem occurs is when a show gets tracked and each track is batch processed as a discrete stream sequence.  The sliding window sees a discontinuity - most likely a step impulse - at the beginning and ending of each track.  When the batch processed tracks are concatenated at playback, there is discontinuity in the noise shaping at each track boundary. 

I doubt that the batch conversion process in R8Brain considers this. But in Samplitude this would be possible because segments are sequential on the same track.

One way to getaround this would be to bounce the whole region from start to finish to one file. Resample that file with R8Brain. Then somehow get the new file broken at the same edit points.

 









 

[Lil Kim Jong-Il]

I doubt that the batch conversion process in R8Brain considers this. But in Samplitude this would be possible because segments are sequential on the same track.

One way to getaround this would be to bounce the whole region from start to finish to one file. Resample that file with R8Brain. Then somehow get the new file broken at the same edit points.
 

Yeah, it probably doesn't although somewhere in the back of my mind I think that there was a version of sox that did this.

If I'm using samplitude, I'll just do the conversion in there since I'm looking for ways to minimize my current workflow.   Is r8brain a better processor than samplitude?

[Brian Skalinder]

If I'm using samplitude, I'll just do the conversion in there since I'm looking for ways to minimize my current workflow.   Is r8brain a better processor than samplitude?

A while back, I did some informal testing of my own between R8Brain Free and SAM.  I decided couldn't hear the difference between SAM's top 2 SRC quality options and R8Brain Free, so I stuck with SAM.  Doesn't mean the differences aren't there, of course, just that I couldn't hear them in my application of the tools, on my playback, etc.  Probably worth doing a comp yourself if you're curious.

[live2496]

If I'm using samplitude, I'll just do the conversion in there since I'm looking for ways to minimize my current workflow.   Is r8brain a better processor than samplitude?

In looking at the graphs I see that R8Brain's sweep is a lot cleaner, but the passband graph shows that R8Brain starts filtering at about 18k whereas Samplitude starts at 21kHz. We are really only talking here about what passes through the process that is originally in the 18-21kHz range. And the artifacts are at -130db or -140db. Those frequencies would be masked by other louder sounds even if we could hear that high! Probably it only matters for test signals.

Weiss' Saracon performs really well, with filtering starting somewhere between 20kHz and 21kHz, but the sweep is really clean too.

[Chilly Brioschi]

Saracon-Lite is almost a kilobuck!
I couldn't bear to look at the DSD version, but I want it, deep in my greedy little heart. 
More German audio software that requires Dongles....   


Do you think that Samplitude would be a step-up from Audition?
I'm looking to redo the DAW, and I guess software is going to be a consideration.   

I always filter before a resample, Audition doesn't process without it.
Not sure what the cut-point is, but since I don't hear anywhere near 19K (thanks FM stereo pilot tones!), it matters little to me.
If I'm flat to 18K, probably 16kHz, that's all I'll ever hear, maybe,.... and on a good day.
Strangely, I can appreciate the difference between  16/44.1 and 24/88.2, although I am unsure why.

I haven't made a 24/96 recording in years, but I am thinking about going to it and letting the CD redbook fall by the wayside.
If I wanted to go WMA-Lossless, would I still need to downsample to 44.1?
Just thinking about a way to get the tunes into the car....

[live2496]

Well, I mentioned Saracon because the graphs looked pretty good.   The best value is R8Brain for SRC and the pro version is not too expensive.

re:dongle
Samplitude requires a dongle too if you go for the non SE versions. Your starting post mentioned workflow and Samp is definitely one way to go to get things done quickly. The SE version is only 50 euros, but it doesn't burn cd's. You have to get into the higher versions for that. I bought the pro version of v.7 a few years ago, so an occasional upgrade keeps me up to date.

I don't know how to get higher than CD quality into a car. I have always considered cars too noisy to hear subtle details. However, I guess you can hear most anything if you turn it up loud enough.     Can you explain your thinking about WMA-lossless? It looks like it supports up to six channels of 24/96 audio.

[macdaddy]

i like sam, too.

i guess, lil kim, that from a vip you could mixdown the audio into one track; the track markers you had inserted will be preserved, then run r8brain, then export the audio so each cd track is its own file...

i have just stuck with that pow-r3 (or whatever that last one is in the dither selection menu)...

[Chilly Brioschi]

i like sam, too.

i guess, lil kim, that from a vip you could mixdown the audio into one track; the track markers you had inserted will be preserved, then run r8brain, then export the audio so each cd track is its own file...

i have just stuck with that pow-r3 (or whatever that last one is in the dither selection menu)...

Wasn't pow-r3 a plugin from MaxiBit or Logic? or am I confusing names from some dither shoot-out?


About the WMA-Lossless....  the car player will play WMA disks, but I'm not sure at what encoding.
Now I have to burn a WMA disc with a bunch of files of different rates and see what happens

   Academic version of Samplitude for under $250...  I just may have to go it.

[Gutbucket]

re:dongle
Samplitude requires a dongle too if you go for the non SE versions. Your starting post mentioned workflow and Samp is definitely one way to go to get things done quickly. The SE version is only 50 euros, but it doesn't burn cd's. You have to get into the higher versions for that. I bought the pro version of v.7 a few years ago, so an occasional upgrade keeps me up to date.

I bought Samplitude v10 Master edition a few months back for $300.  Basically its similar to the full version but limited to 2 track work with only 4 tracks allowed.  No dongle required but has an option to use one if you want to load the app on multiple machines, supports burning CD & DVD, includes some restoration suite plugins, and supposedly has better dither and SRC algorithms than the SE version.  Take a look at the Samp site for a comparison chart if interested.

[live2496]

Wasn't pow-r3 a plugin from MaxiBit or Logic? or am I confusing names from some dither shoot-out?

According to wikipedia it was developed by Millenia Media, Weiss, Dolby Labs and Z-Systems and has been licensed to various companies.
Samplitude and Samplitude Pro and Sequoia have it.