Clients hear dithering difference...

[Jax]

While I was bouncing ref mixes, the clients I'm working with pointed out that a section of a song with harsh, hi-pitched guitar amp feedback sounded overly shrill, and it hadn't when they were listening to the 24 bit session. I know, it's kind of a contradiction - "how nice do you want your harsh feedback to sound?" - but if it bothers a client, it must be fixed. After being drawn in by this anomaly for a closer listen to the whole song, the song sounded "veiled" to them when dithering was applied, but great when it wasn't. I'm used to my mixes becoming slightly veiled when dithering, but I had never heard a client say it before until last night. I would think the veiling effect would reduce shrillness of the one section, but I guess it didn't.

I answered their questions about what was going on by giving them a basic understanding of dithering. When they understood these two things: if dither wasn't applied and the mix was left at 24 bit, it wouldn't playback on their home systems or in their cars - hence my reason for using it; and when the 24 bit mix landed in the hands of a mastering engineer, better dithering tools would be used and its audible effects would be negligible, they were satisfied... after all, the bounces were only refs.

There's only one thing I can think of that might explain why the shrill section stood out. A dual mono compressor plug-in on the master fader (Pro Tools) was followed by multi-channel dither, rather than multi-mono. Would matching the dual mono compression with multi-mono dither have somehow kept the one section from sounding shrill? I didn't think of this last night. We were all very tired and it was the end of the session. I'm sure I was on autopilot. When I go back, I'll check for myself, but I'd like to get some feedback from my favorite ME's here at Gearslutz. What might be the deal?

[masteringhouse]

Originally Posted by Jax

While I was bouncing ref mixes, the clients I'm working with pointed out that a section of a song with harsh, hi-pitched guitar amp feedback sounded overly shrill, and it hadn't when they were listening to the 24 bit session. I know, it's kind of a contradiction - "how nice do you want your harsh feedback to sound?" - but if it bothers a client, it must be fixed. After being drawn in by this anomaly for a closer listen to the whole song, the song sounded "veiled" to them when dithering was applied, but great when it wasn't. I'm used to my mixes becoming slightly veiled when dithering, but I had never heard a client say it before until last night. I would think the veiling effect would reduce shrillness of the one section, but I guess it didn't.

I answered their questions about what was going on by giving them a basic understanding of dithering. When they understood these two things: if dither wasn't applied and the mix was left at 24 bit, it wouldn't playback on their home systems or in their cars - hence my reason for using it; and when the 24 bit mix landed in the hands of a mastering engineer, better dithering tools would be used and its audible effects would be negligible, they were satisfied... after all, the bounces were only refs.

There's only one thing I can think of that might explain why the shrill section stood out. A dual mono compressor plug-in on the master fader (Pro Tools) was followed by multi-channel dither, rather than multi-mono. Would matching the dual mono compression with multi-mono dither have somehow kept the one section from sounding shrill? I didn't think of this last night. We were all very tired and it was the end of the session. I'm sure I was on autopilot. When I go back, I'll check for myself, but I'd like to get some feedback from my favorite ME's here at Gearslutz. What might be the deal?

I would tend to doubt that it had anything to do with using a multi-mono plug but more possibly the type of noise shaping used for the dither. What type of dither did you use?

[Dos]

Are you sure this wasn't a result of Pro Tools' 'Bounce to disk' function?
My mixes sound different when I 'bounce to disk'.
I couldn't even describe, in words, what that difference is and it's not always worse either, just different.
I prefer to bus all the tracks to an audio track, record and export file.

[beachhunt]

Originally Posted by Jax

... it hadn't when they were listening to the 24 bit session...

Er, if they were comparing the 24-bit session with the 16-bit output file, it seems a lot more likely they would object to the difference between 24 and 16 bit audio than objecting to the dither. Compare truncation and dither, see which sounds worse to them (if time/situation allows).

If I misunderstood the above, then perhaps just try another dither if you have the option. Or maybe they'll actually like truncation after all...

[Cellotron]

I very strongly believe that this is a case of barking up the wrong tree as far as what the culprit might be in terms of why a feedback section might seem harsher on one playback than another. I'd much more than likely peg this as being caused by the listener being in a different spot in the control room causing some frequencies to stand out more (see Why We Believe for further details on this point), - or by monitoring at different output levels on the before/after playbacks than to being caused by requantizing. Harsh frequencies will sound much harsher if listened to slightly louder.

First off - do a null test between the source 24bit file and the requantized 16bit file. If there is anything different between them occuring over the -76dBFs level at the very maximum then this is an indication that something is broken in the DAW app or in the dithering algorithm. If there really isn't any difference between the two except at the -80dBFs and lower level - I'd suggest you rethink your idea that dither is capable of making things suddenly "harsher." Some blind a/b/x tests might be in order also.

Best regards,
Steve Berson

[Ethan Winer]

Originally Posted by Cellotron

If there really isn't any difference between the two except at the -80dBFs and lower level - I'd suggest you rethink your idea that dither is capable of making things suddenly "harsher." Some blind a/b/x tests might be in order also.

Of course I agree, and your point about -80 dB is right on the money. That, and the need for a true blind test.

This is not to say that dither is never audible, but it surely cannot account for something sounding more or less harsh, or full, etc as is often reported.

--Ethan

[Jax]

Originally Posted by masteringhouse

I would tend to doubt that it had anything to do with using a multi-mono plug but more possibly the type of noise shaping used for the dither. What type of dither did you use?

Thanks for the replies, all.

I used Pow-R type 3. When I've auditioned the 3 different types in the past, type 3 always seemed the least destructive to fidelity, although its effect is audible. I often use Digidesign dither without noise shaping for quick refs.

Originally Posted by Dos

Are you sure this wasn't a result of Pro Tools' 'Bounce to disk' function?
My mixes sound different when I 'bounce to disk'.
I couldn't even describe, in words, what that difference is and it's not always worse either, just different.
I prefer to bus all the tracks to an audio track, record and export file.

Same here. When I'm sending out files for mastering, I record to a stereo track instead of bouncing to disk. There are huge threads about bouncing to disk, so I don't want to open that can of worms here.

I haven't had the chance to rule out suspects or identify the culprit. The next time I meet the clients is this coming weekend, so I'll have time to experiment and re-examine my past conclusions about dither.

Originally Posted by beachhunt

Er, if they were comparing the 24-bit session with the 16-bit output file, it seems a lot more likely they would object to the difference between 24 and 16 bit audio than objecting to the dither. Compare truncation and dither, see which sounds worse to them (if time/situation allows).

That is worth checking into. Even if they weren't planning to send the files to a ME, I wouldn't want to give them mixes that they hear as compromised. Not every client has the budget for the final step of taking it to a true ME. For those who don't and mention hearing their creations become somewhat degraded, I'd like to get this question resolved and not have to think about it in the future.

Originally Posted by beachhunt

If I misunderstood the above, then perhaps just try another dither if you have the option. Or maybe they'll actually like truncation after all...

On previous rough mixes for them, I used a different dither (Digidesign's) with noise shaping off. There were no comments about mixes sounding worse. I think they're sweating the final details as we near completion, which is good. No other clients have ever commented on these matters before.

[brethes@mac.com]

Originally Posted by Jax

Thanks for the replies, all.

I used Pow-R type 3. When I've auditioned the 3 different types in the past, type 3 always seemed the least destructive to fidelity, although its effect is audible. I often use Digidesign dither without noise shaping for quick refs.
.

In Pro Tools, Pow-R type 3 adds a noticeable brightness. Type 1 or 2 or Digi Dither without noise shaping are more flat.

[Jax]

Steve, it could well be that playback levels were increased, and / or where they sat in the room changed while listening to different mixes. I don't remember if I cranked the level or it remained constant. There's a good chance I did crank it. Anyway, in most listening environments, people don't remain completely idle and listen from the same spot at the same volume every time. A mix should translate in every way possible and related to real life listening.

However, the AE side of me always wants to know why things sometimes work differently than expected. A null test between the 16 bit bounce in question and the 24 bit mix is a good idea, although I'd never point to its result and tell a client that what he hears is wrong. It is only to satisfy my own curiosity and heal anything broken if needed.

Another thing I have to question:

Without playback, I've noticed tiny amounts of movement on the master fader meters when using Pow-R type 3. Whatever it is, it appears to peak @ -75dBFS to -80. Am I watching the intended dither noise in action, or is this NOT normal? I always took it to be normal, but perhaps it could be one of the forces at work here.

[Cellotron]

Originally Posted by brethes@mac.com

In Pro Tools, Pow-R type 3 adds a noticeable brightness. Type 1 or 2 or Digi Dither without noise shaping are more flat.

Brethes -
I need to ask:
Are you stating when requantizing a file to 16bit in PT using POW-R type3 there is additional brightness in elements that peak above -78dBFs?? Actually confirmed as appearing above the 13th least significant bit level in the 16bit file via a null test?

Best regards,
Steve Berson

[Jax]

Originally Posted by Cellotron

I'd suggest you rethink your idea that dither is capable of making things suddenly "harsher."

Fwiw, the idea that dither made a mix element harsher came from my clients, not me. I'm here to investigate their finding.

[Jax]

While Pow-R is up for discussion, here's a reminder of how its designers define the function of each dither type.

[brethes@mac.com]

Originally Posted by Cellotron

Brethes -
I need to ask:
Are you stating when requantizing a file to 16bit in PT using POW-R type3 there is additional brightness in elements that peak above -78dBFs?? Actually confirmed as appearing above the 13th least significant bit level in the 16bit file via a null test?

Best regards,
Steve Berson

Hi Steve,

Sorry, but reading your question is giving me a headache (busy morning). I guess I am not technical enough to make sense of what you are asking.

I am just a mastering engineer, using his ears and hearing a noticeable brightness lift when selecting POW-R type 3 in Pro Tools. Type 1 & 2 and IDR (no noise shaping) don't do this. I had foolishly assumed this was a well know fact (but maybe this only applies to the Pro Tools version).

I can hear the same effect every time so that's good enough proof for me. As mentioned in a previous post, my way of monitoring is different from most of what you guys do as I tend to use a 16bit monitoring chain to make decisions about dithering rather our 24 bit Mytek DA converter as in my view this is closer to the way the CD will sound on most domestic systems (the same way I also check material on NS10s for problems not always obvious on our larger PMCs & Dynaudios).

I rarely choose type 3 because of this HF lift, but occasionally like what it does to a specific material and will choose it rather than going back and tweaking the top end of my EQs.

[masteringhouse]

Originally Posted by brethes@mac.com

In Pro Tools, Pow-R type 3 adds a noticeable brightness. Type 1 or 2 or Digi Dither without noise shaping are more flat.

Agreed, and I wouldn't restrict this to just Pro Tools.

For the feedback issue I would suggest trying type 2, slightly warmer. If your clients agree that this is the issue, and the problem doesn't turn out to be one of the others suggested, I tip my hat to them (if I wore a hat).

[aivoryuk]

Originally Posted by masteringhouse

Agreed, and I wouldn't restrict this to just Pro Tools.

For the feedback issue I would suggest trying type 2, slightly warmer. If your clients agree that this is the issue, and the problem doesn't turn out to be one of the others suggested, I tip my hat to them (if I wore a hat).

yeah i have the same thing in samplitude, in fact i cut a track yesterday where i had used pow-r 3 and did increase the brightness of the the track and altered the sound of the snare quite a bit. I went back to the session that was still open and then just exported it using pow-r 1 and it sounded right

[Cellotron]

Originally Posted by brethes@mac.com

Hi Steve,

Sorry, but reading your question is giving me a headache (busy morning). I guess I am not technical enough to make sense of what you are asking.

A "null test" is where you line up two digital audio files with the same sample accurate start times and then flip the phase of one of the files and sum them. The resulting file shows you what the difference between the two files are.

Dither is generally simply (mostly) random noise. In the case of a noise shaped dither (like POW-R type3) the area with the highest amplitudes of the noise is pushed towards much higher frequencies which tend to be less audible when monitored at normal listening levels. In general the energy of the dither being added when requantizing a file to 16bit does not peak above the 3rd least significant bit (-78dBFs) at the maximum.

So - the questions to be answered with the null test is:
* does the POW-R type 3 dither apply energy well above the noise floor? (in my experience the answer is no - but it would be good to verify that the version being used is not "broken")
* when PT applies the POW-R type 3 dither when requantizing does it create differences from the source file that appear well above the level of the noise floor??? (I don't have direct experience with this at all, but I would guess the answer is no unless PT's handling of POW-R type 3 is also "broken" or if there is some other process being unknowingly applied)
* or is the claim being made that additional frequencies only applied to slightly above the noise floor distinctly change perception of the overall brightness of a file???

If it is the case of the latter then I'd say it would be important to run blind a/b tests to determine that the perception is indeed consistent and not just being "heard" because of expectation or other reason.

I am just a mastering engineer, using his ears and hearing a noticeable brightness lift when selecting POW-R type 3 in Pro Tools. Type 1 & 2 and IDR (no noise shaping) don't do this. I had foolishly assumed this was a well know fact (but maybe this only applies to the Pro Tools version).

OK - that's fine - BUT if we're going to use the word "engineer" in our descriptions of ourselves then that generally indicates that one is willing to do the appropriate amount of research and use scientific methodology in order to gain an understanding of at least the basics of what we are involved with.

So - first off I do not debate that you are perceiving differences - in this case a "brightness" consistently imparted to the requantized file when you are using POW-R type 3 in PT. The question that I'm interested in answering is whether there are truly additional high frequencies being added in these cases that actually peak anywhere over -78dBFs!!

I can hear the same effect every time so that's good enough proof for me.

Yet when things don't logically make sense to me in terms of what a process is actually doing versus the perceptions of the sound after that process has been made I'm very curious as to what the actual reasons are for this.

As mentioned in a previous post, my way of monitoring is different from most of what you guys do as I tend to use a 16bit monitoring chain to make decisions about dithering rather our 24 bit Mytek DA converter as in my view this is closer to the way the CD will sound on most domestic systems (the same way I also check material on NS10s for problems not always obvious on our larger PMCs & Dynaudios).

The DAW system I use (SAWStudio) also can easily be changed with a couple mouse clicks to either do real time output at 24bit, truncated 16bit, or 16bit with a variety of native or plugin dither choices.

I rarely choose type 3 because of this HF lift, but occasionally like what it does to a specific material and will choose it rather than going back and tweaking the top end of my EQs.

That's all well and good - but again - it would be good to use simple scientific experimentation to verify that it is indeed the dither and not something else which is causing the perception of additional brightness after POW-R type 3 has been applied in the DAW app of choice.

Seriously - there has been a number of posters expressing the same things on this thread - will one of you PLEASE perform the null test and let us know what you find? Thanks in advance to whoever is willing to take the time to do this.

Just to clarify - I do not debate that differences in dither types (including their brightness) are not easily perceivable when cranking up fades and tails at maximum levels - but at normal listening levels my experiences (including a few tests of self described "golden ears" in this regard also) is dither choices are more often than not immediately perceivable in any manner close to being ascribed by many in this thread.

Best regards,
Steve Berson

[lucey]

Wow, this is too funny. Talk about a bias, all this theory about how meaningless dither is ... is meaningless to me. I hear it, many do. Could it be that the theory and data on it's insignificance is getting in the way of listening for it? Why don't we stop trying to tell other people what they hear ... what's the point of that exactly? Dither and cable threads should be banned! What good are they.

POWr-3 as I know it, in Samplitude ... is bright. Period. In any material, it's bright/thin. In the chorus, and the verse, not just the tails. POWr-1 is what I've been using lately, sometimes linear. They are different.

The fact that the changes in 16 bit material from dither are -75 or so down doesn't mean they are not audible across the music, or unimportant. Small degrees of low level difference accounts for a huge market in AD DA converters. Dither matters, as does everything audible, and my clients in rare cases have been offered the choice, heard it, and chosen linear or POWr-1 over the other.

POWr-3 I never, ever use.

[Sunbreak Music]

The first time I "heard" dither was w/ TCs X3 and X5, and it was definitely brighter, and unmistakable.

I just figured I was hearing what I paid for.

[Bob Olhsson]

I would avoid multi-mono dither because you don't want it to be correlated between the channels.

[dannygold]

Originally Posted by Dos

Are you sure this wasn't a result of Pro Tools' 'Bounce to disk' function?
My mixes sound different when I 'bounce to disk'.
I couldn't even describe, in words, what that difference is and it's not always worse either, just different.
I prefer to bus all the tracks to an audio track, record and export file.

If you phase invert your bussed recording compared to pro tools bounce, do they cancel out?

[Boomshanka]

Originally Posted by Bob Olhsson

I would avoid multi-mono dither because you don't want it to be correlated between the channels.

+1
And I'm wondering how inadvertently common this might be.

[Jax]

Originally Posted by Jax

Another thing I have to question:

Without playback, I've noticed tiny amounts of movement on the master fader meters when using Pow-R type 3. Whatever it is, it appears to peak @ -75dBFS to -80. Am I watching the intended dither noise in action, or is this NOT normal? I always took it to be normal, but perhaps it could be one of the forces at work here.

Anybody want to tackle this? BTW, the meters are tickled at this very low level only when using Pow-R type 3.

Discussions of dither are extremely fascinating, but I'm trying to resolve an issue here. Thanks.

[masteringhouse]

Originally Posted by Jax

Anybody want to tackle this? BTW, the meters are tickled at this very low level only when using Pow-R type 3.

Discussions of dither are extremely fascinating, but I'm trying to resolve an issue here. Thanks.

What is the noise level without dither and with the other dithers?

[chester]

what about using crane songs dither?

Crane Song Website

[UnderTow]

I did a quick test in Sonar 6. Starting with a 64 bit export of a mix, I exported this file to 16 bits using various dither settings. I used triangular dither as reference and nulled all the other exports to this. These are the resulting maximum sample values:

No Dither: -90.31 dBfs

POW-R1: -73.41 dBfs
POW-R2: -76.33 dBfs
POW-R3: -63.86 dBfs

PSP-FLAT: -84.29 dBfs
PSP-TYPE A: -84.29 dBfs
PSP-TYPE C: -84.29 dBfs

Oxford TPFD: -42.60 dBfs
Oxford TYPE1: -38.99 dBfs
Oxford TYPE4: -42.63 dBfs

L2 TYPE1 NORMAL: -59.94 dBfs

ELEPHANT FLAT: -84.29 dBfs
ELEPHANT SHAPED: -67.39 dbfs


Alistair

[Jax]

Originally Posted by masteringhouse

What is the noise level without dither and with the other dithers?

In the past, I've looked closely at the meters with each type of Pow-R applied, and only type 3 shows any noise at all. There won't be any numbers I can measure before the start of next week, and the visual resolution of Pro Tools meters isn't fine enough to get exact numbers. I'll use Inspector.

There is at least one instance of an intentionally noisy (tape delay) plug-in - massey's TD5, used as the delay on guitar tracks. With TD5 in the mix, the noise level without dither will already be fairly high, but I don't care because the plug-in sounds great. Anyway, the delay on those tracks is muted when the guitar is not playing so there is really more than one noise reference.

What do you expect to ascertain from knowing the noise level? Thanks.

[chrismack]

hmmm... maybe it is not dither.... but something else coinciding... just a guess....

what about intersample clipping.... funny things can happen at nice fractional units of the nyquist frequency; guitar stuff being into 5kHz+ which is near 1/8 Nyquist (and approximately 1/4 of the audio band)...

Have you tried reducing the digital output -3 dB to the DAC (however, your setup is) and making up the gain in analog on the monitors as a potential test?


Cheers,
-chris