SRC before, during, or after dithering?

[Sean McCoy]

A client brought in some 24/48 files for mastering today. I don't think I've had anybody bring in a 48k session since the ADAT days! Anyway, I left the files at 48 while mastering in SS, adding Sonoris dither to the 16-bit output file, then did the SRC afterward using R8brain. I could have done the SRC before mastering, or I could have done it simultaneously in SS while printing the finished file using SS's SRC. Any sonic advantage or disadvantage to any of these three SRC options?

I can't think of any, but I'm kinda fried here in the 105-degree heat at the end of a long day.

[Grekim]

A client brought in some 24/48 files for mastering today. I don't think I've had anybody bring in a 48k session since the ADAT days! Anyway, I left the files at 48 while mastering in SS, adding Sonoris dither to the 16-bit output file, then did the SRC afterward using R8brain. I could have done the SRC before mastering, or I could have done it simultaneously in SS while printing the finished file using SS's SRC. Any sonic advantage or disadvantage to any of these three SRC options?

I can't think of any, but I'm kinda fried here in the 105-degree heat at the end of a long day.

Do your sample rate conversion while it is still a 24 bit file. And beware that your peak levels will change slightly after the SRC, so leave some headroom before you SRC. Dither to 16 bits last.

[TotalSonic]

Do your sample rate conversion while it is still a 24 bit file. And beware that your peak levels will change slightly after the SRC, so leave some headroom before you SRC. Dither to 16 bits last.

Yup! What Grekim said!

The dithering will be essentially "broken" when you perform src as most current src routines do their calculations at a higher bit rate - and as Grekim mentioned, levels can also be changed during a sample rate conversion to a subtle degree. So - src before - requantize/add dither after!

fwiw - I get 48kHz, 88.2 and 96kHz files with some definite frequency here - but I usually do some form of analog processing so I usually play out at the received sample rate and then just capture back at 44.1kHz.

Best regards,
Steve Berson

[Sean McCoy]

Okay, so there is a reason not to do SRC last. :o Thanks, guys. How do you feel about doing it all at the same time right from SS?

[TotalSonic]

Okay, so there is a reason not to do SRC last. :o Thanks, guys. How do you feel about doing it all at the same time right from SS?

Well - the question to Bob is whether the native SRC routines occurs before the native dither options being added and requantizing being done - or also before any output channel processes you have loaded post fader if you instead are using dither plugins (such as the Sonoris Dither / Waves IDR / Izotope Ozone, etc.).

I believe it does work this way - but I'd prefer to do more testing and/or get a confirmation from Bob regarding this before making a definite statement on this. It's easy enough to do it as 2 steps though - and I like the option of being able to archive 24bit files as well.

Best regards,
Steve Berson

[Bob L]

Using the BuildMix routine, the dither would be aded by the MT loop routine and then the SRC would be performed as the code is now.

You would have to do the build and SRC without dither, then do a second pass to add dither if you want it done in that order.

Bob L

[Sean McCoy]

Using the BuildMix routine, the dither would be aded by the MT loop routine and then the SRC would be performed as the code is now.

You would have to do the build and SRC without dither, then do a second pass to add dither if you want it done in that order.

Bob L
Would that be the case even if a dither plugin, such as Sonoris, were placed post-fader on the output track, rather than using the SS dither?

[Bob L]

Yes... all MT processing is done within the loop, then the final data is samplerate converted if needed.

Bob L

[TotalSonic]

Yes... all MT processing is done within the loop, then the final data is samplerate converted if needed.

Bob L

Bob -
Does the native SRC routine occur before the FX Final Res insert though? If it was then that might be an immediate solution for those wanting to do "one pass" src and then requantization/dithering.

Best regards,
Steve Berson

[Bob L]

If you change the rate of the MT itself, then the realtime front-end SRC comes into play as the source data is read from the drive.... but you are then rate converting every track of the MT and ganging up on any artifacts.

If you use the buildmix SRC... that happens after the MT loop as mentioned above... but only one track worth of srtifacts.

Bob L

[TotalSonic]

If you change the rate of the MT itself, then the realtime front-end SRC comes into play as the source data is read from the drive.... but you are then rate converting every track of the MT and ganging up on any artifacts.

If you use the buildmix SRC... that happens after the MT loop as mentioned above... but only one track worth of srtifacts.

Bob L

Bob -
Does the real time front-end src routine go in effect as soon as the MT is set to a different sample rate than the source files even when a "Build Mix" is chosen?

i.e.
When doing a "Build Mix" using highest quality setting chosen in the "build to mix src quality" with source files for this example of 88.2kHz - does the quality of the src change when setting the destination of files to 44.1kHz if the Multitrack Resolution is set at 88.2kHz or 44.1kHz?

Best regards,
Steve Berson

[Bob L]

Steve... any file on the MT that is different than the MT Rate will use the SRC routine and its realtime quality settings... which can be the exact highest setting as the BuildMix... this routine happens at the start of the MT loop right as the source data is read from the drive... so if you set the MT to 88.2k and the files are the same, nothing happens here... but if you force the MT to 44.1k, then every 88.2k file will be converted as it comes off the disk... therefore the SRC happens before any finished dither algorithm... but it happens on each track separately, then the tracks are summed together.

Keeping the MT at 88.2k keeps each file and track intact... sums and processes everything at the 88.2k rate and then converts the finished stereo mix file after the entire MT loop... meaning after any applied dither or other processing within the loop... but, then you are only rate converting one time... not once for each track.

Bob L

[Sean McCoy]

Steve... any file on the MT that is different than the MT Rate will use the SRC routine and its realtime quality settings... which can be the exact highest setting as the BuildMix... this routine happens at the start of the MT loop right as the source data is read from the drive... so if you set the MT to 88.2k and the files are the same, nothing happens here... but if you force the MT to 44.1k, then every 88.2k file will be converted as it comes off the disk... therefore the SRC happens before any finished dither algorithm... but it happens on each track separately, then the tracks are summed together.

Keeping the MT at 88.2k keeps each file and track intact... sums and processes everything at the 88.2k rate and then converts the finished stereo mix file after the entire MT loop... meaning after any applied dither or other processing within the loop... but, then you are only rate converting one time... not once for each track.

Bob L
I'm getting a little lost here. Would this have any effect on a mastering EDL, where there is usually only a single file playing at a time? And even if it did, would this affect quality, or just speed?

[Grekim]

For mastering, I would set the MT rate to what the source files are. Do my eq'ing and compression and make a 24 bit buildmix of that, still at the source sample rate. Then, I'd open a new edl and I change the MT rate to 44.1/24 and do my limiting, now working with 44.1 rate data. Then buildmix with dither and keep the 44.1 kHz, but 16 bits.

[TotalSonic]

For mastering, I would set the MT rate to what the source files are. Do my eq'ing and compression and make a 24 bit buildmix of that, still at the source sample rate. Then, I'd open a new edl and I change the MT rate to 44.1/24 and do my limiting, now working with 44.1 rate data. Then buildmix with dither and keep the 44.1 kHz, but 16 bits.

I should note that many digital peak limiters can often provide results with less artifacts when they are processing at the 2xFs (or above) rates - as this allows them to have smoother anti-aliasing filters for their processing. So in general I would recommend if you are originating from 2xFs (i.e. 88.2kHz or 96kHz) rates that you should do the limiting at this rate first - and leave enough headroom so that no overs will be created by the src and adding dither (generally somwhere from -0.5dBFs to -0.3dBFs works fine for this) - and then src after limiting. And as noted before in this thread - after src - then add dither/requantize.

Best regards,
Steve Berson

[Grekim]

I should note that many digital peak limiters can often provide results with less artifacts when they are processing at the 2xFs (or above) rates - as this allows them to have smoother anti-aliasing filters for their processing. So in general I would recommend if you are originating from 2xFs (i.e. 88.2kHz or 96kHz) rates that you should do the limiting at this rate first - and leave enough headroom so that no overs will be created by the src and adding dither (generally somwhere from -0.5dBFs to -0.3dBFs works fine for this) - and then src after limiting. And as noted before in this thread - after src - then add dither/requantize.

Best regards,
Steve Berson

Hey Steve...I totally agree. When time is not of the essence, I do limiting before SRC. Then play the entire song back through to recheck for overs. Often I need to redo the limiting. A question for you....have you ever known dither alone to actually push a peak to an over? Seems like it shouldn't with the dither level being below -80 dBFs or whatever.

[Dave Labrecque]

Hey Steve...I totally agree. When time is not of the essence, I do limiting before SRC. Then play the entire song back through to recheck for overs. Often I need to redo the limiting. A question for you....have you ever known dither alone to actually push a peak to an over? Seems like it shouldn't with the dither level being below -80 dBFs or whatever.

Depends how close to the top-of-scale you are before adding dither. If you're right to the bleeding edge, then any signal on top of that (e.g., dither) will push you over. Or am I misunderstanding?

I am not Steve Berson. Nor do I play his cello on TV.

[TotalSonic]

Depends how close to the top-of-scale you are before adding dither. If you're right to the bleeding edge, then any signal on top of that (e.g., dither) will push you over. Or am I misunderstanding?

I am not Steve Berson. Nor do I play his cello on TV.

And I am not Dave Labrecque nor do I find the obscure SAW bugs that he somehow always seems to discover!

Anyway - it's indeed very rare that just adding dither will cause an over - BUT as Dave has already pointed out - it is indeed possible for material already going to 0dBFs (and especially if it is at a point where an intersample peak overage might happen anyway) that adding dither can cause an over.

Generally I keep the output ceiling of my masters at -0.3dBFs to prevent problems such as intersample peak overs, or overs resulting from conversions (such as src or using lossy compression odecs such as mp3, etc.).

Best regards,
Steve Berson