Sample Rate Conversion and bit reduction - when and how with software?

[Alex_k]

Hi,

I have some slutty questions about SR conversion and bitrate reduction.

Usually I record tracks of live concerts at 96 KHz / 24 bit. Then I do some mastering in Cubase 4 at the same SR / bits. I also have R8brain pro.

After the editing has been done and I'm going to render the final files for CD burning, I keep wondering what is the best solution between these:

1) apply a dither plug-in on the master and save at 16 bit / 44.1 KHz. So that Cubase will do the Sr conversion also.

2)apply a dither plug-in on the master in Cubase and save at 16 bit / 96 KHz. Then convert the SR with R8Brain.

3) apply a dither plug-in on the master and save at 24 bit / 96 KHz. Convert SR/bits with R8brain. (double dither?!? or no dither on the master? or save 32 bit wavs in Cubase and then apply dither in R8brain only.)

I believe 1 is the worst solution. But I can't decide between 2 and 3.
2 would be faster but I fear to lose something on the bits side.
About 3 I'm unsure about the dither issue.

What do you think? any tips?


Thanks,

Alex

[Cellotron]

Originally Posted by Alex_k

Hi,

I have some slutty questions about SR conversion and bitrate reduction.

Usually I record tracks of live concerts at 96 KHz / 24 bit. Then I do some mastering in Cubase 4 at the same SR / bits. I also have R8brain pro.

After the editing has been done and I'm going to render the final files for CD burning, I keep wondering what is the best solution between these:

1) apply a dither plug-in on the master and save at 16 bit / 44.1 KHz. So that Cubase will do the Sr conversion also.

2)apply a dither plug-in on the master in Cubase and save at 16 bit / 96 KHz. Then convert the SR with R8Brain.

3) apply a dither plug-in on the master and save at 24 bit / 96 KHz. Convert SR/bits with R8brain. (double dither?!? or no dither on the master? or save 32 bit wavs in Cubase and then apply dither in R8brain only.)

I believe 1 is the worst solution. But I can't decide between 2 and 3.
2 would be faster but I fear to lose something on the bits side.
About 3 I'm unsure about the dither issue.

What do you think? any tips?


Thanks,

Alex

You should always do sample rate conversion first prior to requantizating/adding dither - as src algorithms are doing their calculations at higher bit depths than 16.

So -
my suggestions is to
1) sample rate convert from 24bit/96kHz to 24bit/44.1kHz
2) requantize to 16bit/44.1kHz using the dither algorithm of your choice

A nice comparison of sample rate converters can be seen at SRC Comparisons

Best regards,
Steve Berson

[Edward_Vinatea]

Originally Posted by Alex_k

Hi,

I have some slutty questions about SR conversion and bitrate reduction.

Quick "slutty" corrections so that what happened to a friend doesn't happen to you. Bitrate is used for data transfer rates like internet. You want to say 'bit depth' or bit resolution. The word 'Bit' is never in plural {24 bit} so, always bit not bits, although, I have read some very reputable engineers referring it as "bits".

Usually I record tracks of live concerts at 96 KHz / 24 bit. Then I do some mastering in Cubase 4 at the same SR / bits. I also have R8brain pro.


1) apply a dither plug-in on the master and save at 16 bit / 44.1 KHz. So that Cubase will do the Sr conversion also.

Use dither whenever you convert word length, i.e. 24 to 16. Therefore, use option 1. There is no reason to believe that Cubase can't handle the conversion from 24-96 to 16-44.1kHz. Unless you have a hi quality converter which you didn't mention.

Regards,

[Alex_k]

Originally Posted by Edward_Vinatea

Bitrate is used for data transfer rates like internet. You want to say 'bit depth' or bit resolution.

Sorry, I'm a bit-tired today.
Good point!

Originally Posted by Edward_Vinatea

Use dither whenever you convert word length, i.e. 24 to 16. Therefore, use option 1. There is no reason to believe that Cubase can't handle the conversion from 24-96 to 16-44.1kHz. Unless you have a hi quality converter which you didn't mention.

I have tried that but I'm not satisfied with the results. I hear a huge difference between the original tracks played at 96 KHz in Cubase and the rendered files with SRC to 44.1 done in Cubase. In my opinion, Cubase SRC introduces unpleasant artifacts especially in the high frequencies: like metallic resonances and 'ringing'.

That's why I'm considering another approach.

Also when I saw the SRC comparison test posted by Cellotron (thanks man!), my fears about the SRC quality of Cubase 4 were confirmed.

Best,

Alex

[aeonsound]

i found Ozone's guide to dithering is incredibly useful possibly a bit low-level at first for the pros - but deals with all the principles, and really helped me get my head around dithering. not that i'm suggesting you need to, of course

www.izotope.com/products/audio/ozone/OzoneDitheringGuide.pdf

[soundinista]

Supplementary question to OP's:

I'm about to start mixing down some tracks recorded at a live gig to go on a DVD. I'm just doing the audio part, and the video guys will put it all together and edit it.

They told me they would like to work with audio files at 16 bit (I recorded at 24), so my question is -

I have always understood that dithering should be the last process, and that particularly fades and crossfades should not be performed afterwards. So what's the best solution? No dither? Or are any of the various different dithering programs better than others to use in this case, where further editing will be performed on my finished mixes?

Usually I dither with Sonnox Limiter, but I also have Waves and Ozone MBIT available.

Thanks

[aeonsound]

[edit: talking rubbish!]

however, once you have your 16-bit file you can do with it as you please - edit, mix, fade etc. etc...

[soundinista]

Really? How did I get it so wrong?

Thanks for prompt reply

[aeonsound]

firstly, don't take my word for it!

and secondly, apologies - i assumed you weren't treating dither as a separate process before bit depth conversion...

Originally Posted by ozone guide

For dithering to work, it must be the absolute last edit performed on an audio file, except for the final conversion to 16-bit depth. This means that any effect applied after dithering, even a slight gain adjustment, or a sample-rate conversion, can undermine the positive effects of your dithering.

For this reason, it's important to make sure that no changes are made to your mix between dithering and the conversion to 16-bit CD format, and that nothing changes this file afterwards - it needs to be written to CD exactly as it is, with no added gain adjustments or fades. Some hosts may reserve a special slot for dithering plug-ins to reside in, and are designed so that no fades, mixes, or gain changes can occur after the dithering plug-in is applied.

[Nordenstam]

Originally Posted by soundinista

How did I get it so wrong?

You didn't.

Originally Posted by aeonsound

there should be no fades or crossfades before the dither is applied... so you shouldn't apply a gentle automated fadeout to your 24-bit file and then dither down.

however, once you have your 16-bit file you can do with it as you please - edit, mix, fade etc. etc...

It's the other way around. Like your quote from Ozone states.

[aeonsound]

completely and utterly! :D

Mea culpa - Wikipedia, the free encyclopedia

sorry soundinista, i thought you were saying you were worried the DVD editors wouldn't be able to apply any gain changes to your audio once it had been dithered... and somewhere, somehow, misunderestimified in a failtastic way :D

[Nordenstam]

"Misunderestimified in a failtastic way" is a splendiferious use of language! =)

No worries. It's easy to get these things wrong!

Every time someone says "my fault" in latin, it reminds me of Enigma.

Back on topic:

Originally Posted by Edward_Vinatea

There is no reason to believe that Cubase can't handle the conversion from 24-96 to 16-44.1kHz. Unless you have a hi quality converter which you didn't mention.

The graphs at SRC Comparisons is a reason to believe that cubase isn't good enough. He stated that he has r8brain which is a superior SRC routine. Another good option is the free Sox converter.

Originally Posted by soundinista

Or are any of the various different dithering programs better than others to use in this case, where further editing will be performed on my finished mixes?

The thumb rule is to only use noise shaped dither (mbit etc) as the very last dithering and flat/TPDF dither in all situations where the audio may be further processed.

[bob katz]

Alex, some of your choices are breaking the basic rules of digital audio! In short: Dithering to 16 bit should always be the last step. SRC should be the penultimate step.

To elaborate: If you convert to 16 bits first, then SRC, you're actually expanding the wordlength again. And then proposing to truncate it when you convert that to 16 bits...

BK

Originally Posted by Alex_k

Hi,

I have some slutty questions about SR conversion and bitrate reduction.

Usually I record tracks of live concerts at 96 KHz / 24 bit. Then I do some mastering in Cubase 4 at the same SR / bits. I also have R8brain pro.

After the editing has been done and I'm going to render the final files for CD burning, I keep wondering what is the best solution between these:

1) apply a dither plug-in on the master and save at 16 bit / 44.1 KHz. So that Cubase will do the Sr conversion also.

2)apply a dither plug-in on the master in Cubase and save at 16 bit / 96 KHz. Then convert the SR with R8Brain.

3) apply a dither plug-in on the master and save at 24 bit / 96 KHz. Convert SR/bits with R8brain. (double dither?!? or no dither on the master? or save 32 bit wavs in Cubase and then apply dither in R8brain only.)

I believe 1 is the worst solution. But I can't decide between 2 and 3.
2 would be faster but I fear to lose something on the bits side.
About 3 I'm unsure about the dither issue.

What do you think? any tips?


Thanks,

Alex

[kjg]

Originally Posted by Alex_k

I believe 1 is the worst solution. But I can't decide between 2 and 3.
2 would be faster but I fear to lose something on the bits side.
About 3 I'm unsure about the dither issue.

Alex

Bounce to 96k/24b. Don't let Cubase do the SRC.
Convert, dither and truncate with RBrain Pro.

Dithering to 24 bits before bouncing is optional. The dither noise, or the artifacts introduced by truncating without it, will be at a very, very low level.
Theory says that every bit depth reduction should be dithered though. But in this case it is really pretty much arbitrary.

regards,
kjg

[Edward_Vinatea]

Originally Posted by Lupo

The graphs at SRC Comparisons is a reason to believe that cubase isn't good enough. He stated that he has r8brain which is a superior SRC routine. Another good option is the free Sox converter.

I don't work with Cubase. My bad for assuming that it was quality code. Regards,

[soundinista]

The thumb rule is to only use noise shaped dither (mbit etc) as the very last dithering and flat/TPDF dither in all situations where the audio may be further processed.[/quote]

Thanks mate, I think that's what I came on here to find out.

Sorry for hijacking a thread

And - Aeonsound - no worries, thanks for trying to help me out, and respect for your subsequent humility and self-correction. We should all learn from your example.

You guys rock

[Sweep]

i use Bias Peak w/ the pwr-r dither and it always sounds great......what bob katz said about the truncating is correct.....stick with that process.....btw peak only costs $200 and serves as a great waveform editor as well....grammy winning engineers told me that the pwr-r setting was almost as good as the tools they used

[macc]

Originally Posted by Lupo

The graphs at SRC Comparisons is a reason to believe that cubase isn't good enough. He stated that he has r8brain which is a superior SRC routine. Another good option is the free Sox converter.

Sorry for bumping this thread, but I finally got round to trying out SOX for sample rate conversions and I am really impressed!

Thanks for the tip Lupo

[Adam Dempsey]

Originally Posted by Edward_Vinatea

Bitrate is used for data transfer rates like internet. You want to say 'bit depth' or bit resolution. The word 'Bit' is never in plural {24 bit} so, always bit not bits, although, I have read some very reputable engineers referring it as "bits".

There are 8 bits in a byte.
A sample word can contain 'x' bits...

But yes bitrate = a data rate (eg 320kb/s)
whereas 'x' bit audio refers to wordlength = bit depth/ bit resolution.

[thedigitalgod]

Originally Posted by bob katz

Alex, some of your choices are breaking the basic rules of digital audio! In short: Dithering to 16 bit should always be the last step. SRC should be the penultimate step.

To elaborate: If you convert to 16 bits first, then SRC, you're actually expanding the wordlength again. And then proposing to truncate it when you convert that to 16 bits...

BK

Mr. Katz, if i can clarify, the absolute best case scenario would be:
(recording at 88.2/24)
-mix down to 88.2/24
-convert sample rate to 44.1
-add dithering
-truncate to 16 bit

does it matter if the dithering is added at the mixdown stage or after sample rate conversion to 44.1?

also, the process of "dithering" is always just adding dithering noise and then truncating, correct? there is no difference in "dithering to 16 bit" and "adding dithering and then truncating". . . is that correct?

im being very specific because ive read a ton about this, including everything on your site and my little mind seems to go numb and not be able to formulate a solid opinion.

thanks so much for your time.

-shawn

[Bob Olhsson]

Under the hood all digital audio is numbers. Every signal processing and most mixing steps involve adding, subtracting, multiplying and dividing the numbers. At the very end of the process we need to fit the final number calculated into a limited space. The best sounding way to do this is to randomize the last digit that will fit using a dither process so that the cut-off sounds like random noise rather than distortion.

This always needs to be done as the final step that is committed to a disk or a file.

[bob katz]

Originally Posted by thedigitalgod

Mr. Katz, if i can clarify, the absolute best case scenario would be:
(recording at 88.2/24)
-mix down to 88.2/24
-convert sample rate to 44.1
-add dithering
-truncate to 16 bit

does it matter if the dithering is added at the mixdown stage or after sample rate conversion to 44.1?

You were doing great until your last question! As Bob O. just mentioned, calculations INCREASE wordlength. So it is senseless and debilitating to decrease the wordlength first and then do further calculations or conversion. And in fact when you study just one of the steps in your note above "convert sample rate to 44.1" there is an implicit wordlength increase to the internal wordlength and design of the sample rate converter. So your 24 bit/88.2 mixdown may increase to a 32 bit float/44.1 kHz signal after SRC. And that's good, it means you are using all the internal resolution of your sample rate converter before dithering down to 16 bit.

In summary: Let it grow till the end. Don't shrink the pipeline prematurely.

Yes there are diminishing returns, but this is the proper way to do it.

Hope this helps,


BK

[thedigitalgod]

Originally Posted by bob katz

You were doing great until your last question! As Bob O. just mentioned, calculations INCREASE wordlength. So it is senseless and debilitating to decrease the wordlength first and then do further calculations or conversion. And in fact when you study just one of the steps in your note above "convert sample rate to 44.1" there is an implicit wordlength increase to the internal wordlength and design of the sample rate converter. So your 24 bit/88.2 mixdown may increase to a 32 bit float/44.1 kHz signal after SRC. And that's good, it means you are using all the internal resolution of your sample rate converter before dithering down to 16 bit.

In summary: Let it grow till the end. Don't shrink the pipeline prematurely.

Yes there are diminishing returns, but this is the proper way to do it.

Hope this helps,


BK

thanks so much for both your replies Mr. Olhsson and Mr. Katz, i feel quite lucky to have the both of you directly answering questions.

please forgive me if i seem a little thick, but i just want to make sure i dont misunderstand:

dither then truncate (as it seems Mr. Katz suggested) or truncate then dither (as it seems Mr. Ohlsson suggests)?

and the truncation should be identical in any software used -- wavelab, SoX, whatever; because it simply lops off 8 bits, nothing magical there (please correct me if im wrong)

this is definitely the last quarter of a percent of quality in the grand scheme of things, and ive tested all above-mentioned orders of operation and the differences are obviously quite miniscule to the point that i cannot say "yes, this one sounds better" -- but i would like to know that im doing things the proper way and put this area behind me so i can worry about the larger pieces of the pie.

again, thank you both for your time educating those of us with lesser understanding.

-shawn

[bob katz]

Originally Posted by thedigitalgod

thanks so much for both your replies Mr. Olhsson and Mr. Katz, i feel quite lucky to have the both of you directly answering questions.

please forgive me if i seem a little thick, but i just want to make sure i dont misunderstand:

dither then truncate (as it seems Mr. Katz suggested) or truncate then dither (as it seems Mr. Ohlsson suggests)?

I'm sure my friend Mr. Olhsson (probably the world's greatest typographical error) knows what he's saying and that it's just a matter of interpretation.

The answer is actually academic in that 99.99% of all dithering utilities create a new file at the proper wavelength while dithering. So the proper order of events is taken care of for you. But when in doubt, dither (to the proper amount of dithering), then truncate (to the proper wordlength determined by the amount of dithering that was applied).

For example: You put a 24 bit file into a dithering utility. Out comes a 16 bit file. It's been dithered and truncated for you in that order.

this is definitely the last quarter of a percent of quality in the grand scheme of things, and ive tested all above-mentioned orders of operation and the differences are obviously quite miniscule to the point that i cannot say "yes, this one sounds better" -- but i would like to know that im doing things the proper way and put this area behind me so i can worry about the larger pieces of the pie.

Yes, once you get this behind you, then you can argue with Ethan Winer over how many angels can fit on the head of a pin. Like chicken soup, it couldn't hurt, and it probably helps. The symptoms of improper dithering include: colder sound, loss of depth and space. I can't imagine anyone taking a shortcut on that.

BK

[thedigitalgod]

thank you again Mr. Katz, that certainly cleared it all up for me. of what particular dithering utilities do you speak? all the software utilities i have used simply add the dithering noise but require that the truncation be manual -- for instance nuendo, wavelab, reaper, etc, where dithering is merely a plugin and the specifics of the file saved are up to the user. i suppose, given the quality of gear that you use, you are speaking of hardware units; could you elaborate?

[Audiop]

Above Bob wrote "wavelength while dithering" but I assume that was a typo and it should be "wordlength". :-)

@ thedigitalgod,

In Reaper while you render a files you have the checkboxes for wordlength of choice and also dither + noise shaping. I'm sure things are done in the right order if you choose both truncation and dither.


/Peter

[bob katz]

Originally Posted by thedigitalgod

thank you again Mr. Katz, that certainly cleared it all up for me. of what particular dithering utilities do you speak? all the software utilities i have used simply add the dithering noise but require that the truncation be manual -- for instance nuendo, wavelab, reaper, etc, where dithering is merely a plugin and the specifics of the file saved are up to the user. i suppose, given the quality of gear that you use, you are speaking of hardware units; could you elaborate?

You're right. Most DAWs let you pop in a dithering plugin. And then it your responsibility to choose and truncate the file to the proper length. But if you look at the output of the dithering plugins with a bitscope, you will see that the majority of them truncate the output properly as well. Yamaha used to add 16 bit dither but not truncate so you were hearing the extra bits below 16 not truncated.

As for outboard converters, there are many. Weiss Saracon. R8Brain Pro. AudioEase Barbabatch. Sample Manager Pro. And so on and so on, for both mac and PC. R8Brain does not incorporate choosable dither to my knowledge so I recommend outputting to a 32 bit float file, which to my knowledge is its internal wordlength. And the devices which have a check box for "dither" (such as Barbabatch) will calculate the dither amplitude properly for the output wordlength that you suggest and will truncate it, but really, truncation is as simple as making a file of the proper wordlength and discarding all the extra bits.

Excuse me for using the word "devices" when in most cases we're talking about software utilities. Of course there are outboard hardware sample rate converters and dithering boxes as well, notably from Weiss, and Dan Lavry's 3000 converter as well as some boxes by Prism.

BK

[Bob Olhsson]

NEVER ASSUME audio software developers or audio manufacturers know what they are doing! It simply doesn't pay enough to attract the best engineering talent.

[thedigitalgod]

awesome, thanks so much -- i finally feel like i actually have an understanding of all this rather than just "doing what im supposed to do". your help is greatly appreciated, Mr. Katz. and thanks to you as well Mr. Olhsson.

-shawn

[Tube World]

The Samplitude SRC seems a lot better than Cubase 4 according to that SRC comparrison. I just wonder if I would really hear an audible difference between Samp and RBrain Pro? It's obviously a lot easier to do everything inside Samplitude.