Skip to main content

Mixing & Mastering at 96kHz

Member for

6 years 1 month

Found this an interesting read on mixing & mastering at 96kHz and thought it would be worth sharing.

http://www.ryanschwabe.com/blog/96k

Comments

Member for

6 years 1 month

Sean G Sun, 09/11/2016 - 22:56
What I took from it is that @96k any plug-ins used create cleaner harmonics, @44.1k plug-ins such as compressors and equalisers create aliasing foldback below 44.1k that add clutter to the audio spectrum. The more plug-ins used add to the harmonics which add to the aliasing foldback.

Member for

5 years 1 month

Brother Junk Mon, 09/12/2016 - 05:22
That makes sense. So if we run our stuff 24/96 it moves much of the aliasing (NSF or other) to beyond the point of human hearing. I wonder if any of the 44.1 problems would go away if it was 24/48? That's what i usually run in...I can do 96 or up to 192.

I'd be curious what would happen if you ran it at 192?

Also, are many people using 32 bit float?

I don't quite get the phase inversion part of it. I understand phase, I understand how passives affect phase, how crossover slopes affect phase, how eq's can affect phase. I've worked in environments where in order to get things to sound right, you need to flip the phase of a driver. I perhaps don't understand it as much as most of you, but I have a pretty good handle on it. I had a fully adjustable phase switch for a long time, so I got to play around with phase a lot.





What I don't get is inverting the phase of the track at 44.1 and 96? Or can anyone explain how that phase inversion track is working? I understand what he is proving with the track...I don't understand what he did to make that track?

Here's a better way to ask it, here is a quote from the paragraph above that track...

"Below is a stream of the phase inverted difference between the 96kHz session bounce and the 44.1kHz session bounce."

Can someone explain what he is talking about?

Member for

6 years 1 month

Sean G Mon, 09/12/2016 - 05:48
If you have 2 signals and they are exactly the same, when you invert the phase of one it cancels out so in effect you should have silence.

If the signals are not exactly the same and you invert the phase on one, you hear the difference between the two.

What has been done is one is at 44.1 and the other at 96k so what you are hearing is the difference between the two signals when one is inverted, or what is left behind after what is the same with both signals has been cancelled out.

What you are effectively hearing is the difference between the two files.

Member for

6 years 1 month

Sean G Mon, 09/12/2016 - 05:53

Excerpt from the last paragraph...

"In the above phase inversion test you can clearly hear the aliasing on the open hi-hat. Second, there is a distinct amount of high frequency detail that is prevalent in the 96kHz bounce that is not captured in the same way in the 44.1kHz bounce. This high frequency detail can be heard in what remains of the vocal in the phase inversion test above. Third, higher sample rates allow you to control transient detail with more precision and less distortion than at lower sample rates. This is why we see oversampling features built into many popular digital mastering limiters...."

Member for

5 years 1 month

Brother Junk Mon, 09/12/2016 - 05:59
Sean G, post: 441212, member: 49362 wrote: "In the above phase inversion test you can clearly hear the aliasing on the open hi-hat. Second, there is a distinct amount of high frequency detail that is prevalent in the 96kHz bounce that is not captured in the same way in the 44.1kHz bounce. This high frequency detail can be heard in what remains of the vocal in the phase inversion test above. Third, higher sample rates allow you to control transient detail with more precision and less distortion than at lower sample rates. This is why we see oversampling features built into many popular digital mastering limiters...."
I get that part...your first post pretty much answered my question.

Im gonna have to mess with it myself to answer my other questions. Thank you Sean G.

So do most people here use a 96khz sample rate?

Member for

7 years 2 months

Chris Perra Mon, 09/12/2016 - 21:11
Everything sounds different than everything else.. regardless of what it is.. No 2 Mics. Pres. Convertors,.. Speakers.. Rooms. Performances.. Are exactly the same on a measurement scale.. If it sounds good.. it is.. if it doesn't it probably has alot more to do with than 96k vs 44.1

When I render using real time vs regular in Cubase using the same sample rate they don't null.. to me the real time sounds better..

Also.. all this time I thought the industry was trying to add analog distortion/saturation and character to mixes.. I wonder what the same tone generator tests with analog gear would be like.
Would they have the same less harmonic distortion as 96k or more? less than 44.1 or more?.

I wonder.. If there is a difference.. would the plug in manufacturers try and get the exact modeling to 44.1 ..48..96k.. or 192 because if they are different.. then bench marking would be different as well.. and the correct exact model would have to be at one of those sample rates..

Member for

7 years 2 months

Chris Perra Mon, 09/12/2016 - 21:21
I just read that flow chart.. that's a bizzarre way of doing things.. record in Ableton live then upsample to Protools at 96k .. Do a mix at 96k.. Then copy the 96k mix and then downsample it to 32 bit 44.1 ... render both and the compare them...

Why not do a mix at 96 or 44.1 and then copy the mixer settings to the other using the original source files for both..

Member for

6 years 1 month

Sean G Mon, 09/12/2016 - 21:32
Chris Perra, post: 441230, member: 48232 wrote: I just read that flow chart.. that's a bizzarre way of doing things.. record in Ableton live then upsample to Protools at 96k .. Do a mix at 96k.. Then copy the 96k mix and then downsample it to 32 bit 44.1 ... render both and the compare them...

Why not do a mix at 96 or 44.1 and then copy the mixer settings to the other using the original source files for both..

I thought the process in the flow chart looked at little weird too

Member for

5 years 1 month

Brother Junk Tue, 09/13/2016 - 05:19
pcrecord, post: 441223, member: 46460 wrote: I do track, mix and master in 96khz, sounds better to me ;)
Merci. I've always done 24/48 bc of computer restrictions, but I don't have those now. 96 here I come.
Chris Perra, post: 441229, member: 48232 wrote: I wonder.. If there is a difference.. would the plug in manufacturers try and get the exact modeling to 44.1 ..48..96k.. or 192 because if they are different.. then bench marking would be different as well.. and the correct exact model would have to be at one of those sample rates..
That is actually an interesting thought. I wonder as well now...
Chris Perra, post: 441230, member: 48232 wrote: I just read that flow chart.. that's a bizzarre way of doing things.. record in Ableton live then upsample to Protools at 96k .. Do a mix at 96k.. Then copy the 96k mix and then downsample it to 32 bit 44.1 ... render both and the compare them...

Why not do a mix at 96 or 44.1 and then copy the mixer settings to the other using the original source files for both..
I did not understand that part at all either. I understand what he is saying...I don't understand why anyone would do what he shows.

Please excuse this if it's a stupid question...I never really thought about it. But according to that chart, he is changing the bit depth, and sample rate, often up! 24 to 32, 44.1 to 96.

Once the track is recorded, can you actually change the track? Not on paper, but, actually alter the track? Once it's been recorded in 16/44.1, you can change it to 24/96 to the and it will sound as if you recorded it in 24/96?

I thought past the recording stage, you could change the bit depth and sample rate down...but not up. Can someone tell me which is correct?

Member for

11 years 8 months

bouldersound Tue, 09/13/2016 - 22:36
In Sony Vegas you don't necessarily have to upsample your files, you can just change the project properties to whatever rate you like and it will resample on the fly. I suppose there's a CPU penalty for that, but it does it.

Hey, wait, I though resampling was supposed to be a bad thing. Now all of a sudden it's a good thing?

Member for

15 years 5 months

Boswell Wed, 09/14/2016 - 04:18
The Ryan Schwabe article that Sean linked in the first post of this thread is interesting, but the author falls down on much of what he attempts to find fault with.

First example: take the diagram that shows the spectrum of a waveform whose fundamental is not stated, but presumably at about 7KHz. Quote: "At 44.1kHz sampling rate the 3rd harmonic is below the anti-aliasing cutoff filter, but above the Nyquist-Shannon frequency". If you look at what he has drawn as an anti-aliaising filter, it's only about 5dB down at the Nyquist frequency, so it's simply the wrong frequency of filter to be using. Anti-aliasing filters designed for the CD audio frequency of 44.1KHz are 3dB down at 20KHz and somewhere around 70dB down at the Nyquist. His incorrect example leads to false conclusions.

Second example: the assumption that filters that perform anti-aliaising in the analogue domain ahead of sampling work at the top end of the audio band. In the last 15 - 20 years, the only types of ADC and DAC used for sampling and reconstruction of audio sample well outside the audio frequency band, typically at many MHz. At this rate, it's possible to reduce the wordlength from 16 down to 4 or even 1 bit, and then recreate the longer wordlengths during the decimation (down-sampling) process in the digital domain. The point is that, when sampling at these MHz rates, the analogue anti-aliaising filters can be very simple designs that roll-off gently in the tens or even hundreds of KHz, causing minimal phase irregularities in the audio band. The anti-aliaising filters for containing the information in the band 0 - 20KHz are implemented digitally as part of the decimation process.

Ryan Schwabe is right to raise many of the issues that he does in his article, but he goes about justifying his conclusions in the wrong way.

The further subject of whether DAW plug-ins can generate aliaising products is down to the coders of the individual plug-ins and the structure of the DAWs they are running in. Certainly, a plug-in that intentionally creates distortion (generating harmonics not present in the original waveform, or increasing the amplitude of those that are) lays itself open to problems attributable to the coder's lack of full understanding of the physics involved. A properly designed and implemented algorithm will suppress harmonic components that would otherwise fall above the Nyquist frequency. Note that algorithms of this type do not automatically know what clock rate they are running at; they can only deal in terms of fractions of the sampling frequency, whatever that happens to be at the time of use. If, for example, a DAW allows a plug-in to run at 96KHz in a 44.1KHz session, then it's up to the DAW and not the plug-in to perform SRC at the input and output of the plug-in, together with any anti-aliaising band-limiting associated with conversion to a lower rate. It's a tricky subject, and I've had suspicions for a long time that many DAWs do not implement this correctly, let alone any problems due to the use of digital SRC. I've yet to be persuaded that the benefits of running plug-ins at sampling rates different to that of the session outweigh the problems that the necessary implied SRC can cause.

Member for

6 years 1 month

Sean G Wed, 09/14/2016 - 05:27
Boswell, post: 441244, member: 29034 wrote: If, for example, a DAW allows a plug-in to run at 96KHz in a 44.1KHz session, then it's up to the DAW and not the plug-in to perform SRC at the input and output of the plug-in, together with any anti-aliaising band-limiting associated with conversion to a lower rate. It's a tricky subject, and I've had suspicions for a long time that many DAWs do not implement this correctly, let alone any problems due to the use of digital SRC. I've yet to be persuaded that the benefits of running plug-ins at sampling rates different to that of the session outweigh the problems that the necessary implied SRC can cause.

Thanks for sharing your understanding of this Bos, the issue of SRC and the downside of this is something I came away scratching my head over.

With plug-ins that oversample to perform their given function and the DAW performing the SRC at the input and output stage of the plug-in, say in the case where the session is at 44.1kHz, if the DAW is not implementing this correctly is it safe to say that some aliasing foldback maybe added to the audio spectrum?

Member for

15 years 5 months

Boswell Wed, 09/14/2016 - 05:37
Sean G, post: 441245, member: 49362 wrote: With plug-ins that oversample to perform their given function and the DAW performing the SRC at the input and output stage of the plug-in, say in the case where the session is at 44.1kHz, if the DAW is not implementing this correctly is it safe to say that some aliasing foldback maybe added to the audio spectrum?
Yes, except in the case where a plug-in is specifically written to run at a sampling higher rate than the DAW session. In that case, the responsibility is with the plug-in to perform the required SRC and anti-aliaising at its output. These types should be safe to use with any DAW and not cause foldback, but they may separately still suffer from digital SRC effects (my pet gripe).

Member for

21 years

audiokid Wed, 09/14/2016 - 11:34
Interesting thread, thanks Sean.

44.1/24, 48k/24 sounds very good to me. I do hear an extra sweet smoothness at higher SR but the extra sweetness has never been a game changer; to always be set on example: 96k.
My thinking... when using top quality converters, a win win for both sound quality and DAW performance can be accomplished from lower SR . Budget converters don't seem to have the same results.

Samplitude, PCIe interfacing and a well built PC mutitracks very smooth @44.1 for me, which is why I choose this platform as my "go to" DAW as well as why good converters are worth the investment.

Member for

8 years 7 months

pcrecord Wed, 09/14/2016 - 12:01
My modo is; everytime you process audio, you degrade it. Some degradation are a good thing, like when going for analog external gear, but ITB it's bad. I don't thrust computers to do the job perfectly each time. (Yes I'm an IT ;))
If you record at 44, mix at 44. I could even say master at 44 as well because most high end plugins have oversampling anyway so the excuse that they will sound better drops right there... I haven't always thought about it this way, but this is my actual opinion ;)
x