Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[bobkatz]

Since we're also talking about audiophile quality improvements, I'd like to see proper TPDF dithering, rather than the less-adequate dithering  which is currently implemented. And I'm talking about the dithering to 24 bit.

Here is the loss I notice:

a) MC playing 24 bit audio when the volume control is off of the zero point, feeding a 24-bit DAC directly, versus

b) MC with a 24-bit dithering plugin. Sounds better. Unfortunately, multichannel plugins don't seem to work right, so I cannot use this solution in 5.1

or

c) MC running 64-bit float, Acourate ASIO into Acourate Convolver. Acourate Convolver is properly dithered and has more apparent depth and resolution than MC doing its own dither.

In both cases, to my ears b and c sound better (warmer and deeper) than a. I can only conclude that the dither in MC is not an accurately-implemented TPDF.

Please don't ask me to pass a blind test on this. This is so subtle that it's very difficult to pass blind, guys.

On the Mac side, when I turn off dither in the advanced preferences, and implement a plugin for dither, I get the same sonic improvements as on the PC.

[bobkatz]

I am aware of the discussion and participated in it. I am a recognized authority on this subject. In a portion of the discussion I actually helped Matt get out of a bind by analyzing JRiver's 24 bit dither and concluding (at that time) that it was certainly satisfactory at 24 bits despite there being a bit of visible noise modulation on an FFT analyzer.

Subsequently (not reported in this forum) I did further listening and investigation and came to the conclusion that JRiver's 24 bit dither is audibly "good but not perfect".

The reasons (I repeat) that I did not decide to discuss that at that time were that yes, the topic had gotten ad nauseum and a rather rude individual who wanted to see a more perfect 16-bit reduction was just getting on our nerves. In addition, a good dithering plugin corrects the replaces and corrects the slightly weaker resolution of JRiver. So I let it go.

But now that JRiver 21 is coming in and JRiver has reached a high level of maturity, I decided to readdress the topic. Audiophile-quality processing is the goal of JRiver and if it does not perform audibly as well as a textbook-implemented random TPDF then it should do so. TPDF with a proper random number generator is about 20 lines of C++ .

[dallasjustice]

I really can't say whether one dither is better than another. I can say that I've compared Acourate Convolver's TPDF dither with it's shaped either. I honestly can't hear any difference.

Bob, what perceptual cues do you hear or not hear with JRiver's dither which you can/can't hear with other dither?  I know your ear is trained for this and mine is not, so I defer to you.

Michael.

[bobkatz]

I really can't say whether one dither is better than another. I can say that I've compared Acourate Convolver's TPDF dither with it's shaped either. I honestly can't hear any difference.

Bob, what perceptual cues do you hear or not hear with JRiver's dither which you can/can't hear with other dither?  I know your ear is trained for this and mine is not, so I defer to you.

Michael.

Dear Michael: It sounds like a very subtle loss of depth (ambience) to me. This is not earth-shattering, but there is only one correct TPDF. I can't hear a difference between Acourate Convolver's shapes at 24 bit either! All the shapes start as flat TPDF, which is the important point. The shaping is really only helpful at lesser bit-depths.

Also keep in mind that other well-implemented dithers as plugins when inserted into JRiver sound better to me than JRiver's current dither.

This is an "audiophile level difference" folks, it's not earth-shattering, I could not pass a blind test on this, but I believe it is audible. It is certainly measurable as a bit of noise modulation in an FFT, which is a symptom of inadequately-random noise generation.

[DarkPenguin]

I could not pass a blind test on this, but I believe it is audible.

I don't get what you mean by this.

[JimH]

I don't get what you mean by this.

Don't fall into this trap, Bob. 

[AndrewFG]

Don't fall into this trap, Bob. 

Jim, you may want to change the subject of this thread from "Better Dithering" to "Alternative Dithering" ..

[JimH]

Bob thinks it's better.  I'll change it though.

[DarkPenguin]

Don't fall into this trap, Bob. 

It is a legitimate question!  I don't know if he means it isn't always audible.  It isn't consistently better.  Or what exactly.

I should think you could ferret this out with a long double blind and then analyzing whether or not a difference was heard when there was in fact a difference to be heard.

[Hendrik]

We'll look into this. Likeliest choice is that we'll add a option to change the behavior for the time being, instead of changing the default.

[dallasjustice]

Bob's account seems plausible to me.

I used to own a totaldac. I had a friend who said he could hear the difference between the volume control in jriver and the totaldac's volume control. I put him to the test and he could consistently pick out the Totaldac volume control in a blind test. He described the difference in terms of spatial cues; eg. Width and depth. When I did the test I disabled the jriver volume and used the totalDAC internal dithered volume control. He preferred the totaldac every time. Personally, I couldn't (or didn't want to) hear the difference. :-)

[Hendrik]

The next build will offer full-strength TPDF dithering (ie 2LSB as people like to call it), but not enabled by default. There is a new option in Audio -> Advanced -> Dither Mode

If anyone is interested in seeing a difference, even if you cannot hear it, I did test with the Bit-depth simulator plugin in MCs PEQ set to 16-bit, and the Voxengo SPAN visualizer VST (dial the range all the way down to -140dB). You can see low distortions with our original dithering, and the perfectly even, but stronger, noise with TPDF.

[stewart_pk]

Hi, I'm using a MF V-Link 192 with it's supplied Thesycon ASIO driver.
The driver appears to only accept data at 32bit but I'm not 100% sure of this.
The MF-V-Link only outputs data at 24 bit to a DAC so I assume it may dither down internally from 32 bit to 24 bit rather than truncating.
I use JRiver's Internal volume. Should I enable or disable JRiver's dither option?

[Hendrik]

You should always dither.

[stewart_pk]

Hi, thanks for your quick reply. I don't fully understand why I should always dither but you're the expert not me so dither it is

Another question if you don't mind (or somebody else). If I use ASIO then Audio Path reports 32 bit output. WASAPI on the other hand with Automatic Bit Depth set it in the WASAPI options then Audio Path reports 24 bit (padded) output. I appreciate there may be no audible difference especially in real life use but which one is technically superior?

[mwillems]

The next build will offer full-strength TPDF dithering (ie 2LSB as people like to call it), but not enabled by default. There is a new option in Audio -> Advanced -> Dither Mode

If anyone is interested in seeing a difference, even if you cannot hear it, I did test with the Bit-depth simulator plugin in MCs PEQ set to 16-bit, and the Voxengo SPAN visualizer VST (dial the range all the way down to -140dB). You can see low distortions with our original dithering, and the perfectly even, but stronger, noise with TPDF.

I'm glad to hear this.  It will make some folks very happy, and hopefully end a long-standing source of forum angst.  Let me know if you want measurements or testing (although I suspect Bob will probably have that covered in spades).

[Hilton]

Wow... thanks Bobkatz and Thanks Hendrik for listening. LOL

I'd never of guessed this would make it into a build.

[bobkatz]

The next build will offer full-strength TPDF dithering (ie 2LSB as people like to call it), but not enabled by default. There is a new option in Audio -> Advanced -> Dither Mode

If anyone is interested in seeing a difference, even if you cannot hear it, I did test with the Bit-depth simulator plugin in MCs PEQ set to 16-bit, and the Voxengo SPAN visualizer VST (dial the range all the way down to -140dB). You can see low distortions with our original dithering, and the perfectly even, but stronger, noise with TPDF.

Dear Hendrik: To me that's fantastic news. As soon as that build is announced I'd like to test it. As far as blind test goes, what I meant is that I notice the results sighted.

Basically, I claim to hear a difference that's very subtle and difficult to prove in a well-organized blind test. It should be acknowledged that blind tests are very fatiguing and therefore subtle differences which are claimed to be audible sighted may not be easily provable. And some assertions have to be taken as a positive indication, as you have. It's how audio advanced from tubes to transistors and back to tubes, isn't it  :-).

It is very difficult to organize a formal blind test on such a subtle difference. It could take weeks to months of painstaking trials to prove this assertion. In fact, it takes a long long long time to prove any assertion of subtle difference blind. Would the folks in JRiver undertake to improve their video resolution if they learned the vast majority of the public can't tell the difference between 2K and 4K video?

As my colleague Dave Collins says, "If it's done blind, it's amazing you can tell what piece of music is playing". Well, it's not that bad, but anyone who has tried to organize a serious blind test knows what I'm talking about. And there is no such thing as a "casual blind test", no sir. Don't fall into that trap. Results are likely to be random with an uncontrolled, "casual" blind test. I'm conducting a formal blind test right now that's making me pull my hair out!

[CountryBumkin]

I'm glad to hear this.  It will make some folks very happy, and hopefully end a long-standing source of forum angst.

+1

[kstuart]

If I am understanding this issue correctly, if one has Volume in MC21 set to "Disabled Volume", then the dither is never used.

Is that correct ?

[mojave]

If I am understanding this issue correctly, if one has Volume in MC21 set to "Disabled Volume", then the dither is never used.

Is that correct ?

DSP is all done at 64-bits in JRiver. Anytime there has been any DSP done it is dithered when converted back to either 32 or 24 bits for the DAC. This includes Parametric Equalizer, Volume Leveling, Adaptive Volume, etc.

[v_erich]

+1

Thanks

[kstuart]

DSP is all done at 64-bits in JRiver. Anytime there has been any DSP done it is dithered when converted back to either 32 or 24 bits for the DAC. This includes Parametric Equalizer, Volume Leveling, Adaptive Volume, etc.

If Volume is set to "disabled volume" and the Audio Path is "blue", then no dithering at all is used.

Is that correct ?

[mwillems]

If Volume is set to "disabled volume" and the Audio Path is "blue", then no dithering at all is used.

Is that correct ?

No. Some volume changes (such as those done by volume leveling or adaptive volume) will not turn off the "blue light", and you'd be getting dither under those circumstances.  The only way to be certain no dithering of any kind is happening is to disable volume, use a direct playback method that bypasses the windows mixer, and make sure that all JRiver DSP blocks are disabled.

[Hendrik]

You'll probably also want to disable dithering if you're trying to go for that, otherwise there are no guarantees.
But who plays without any kind of DSP anyway, and if its only loudness and volume leveling.

[mojave]

I used Tools > Advanced Tools > Audio Calibration > Tones to output a 1000 Hz sine wave. I played at -14 dB. The noise of the sine wave didn't change with any volume change. Only the actual sine wave peak level changed.

I viewed it using Voxengo Span in JRiver's DSP Studio. Here are the results with 64 bit output, 16 bit output with no dither, 16 bit output with JRiver dither and 16 bit output with TPDF dither.

[Hendrik]

I don't think your No Dither image is actually correct. Or I broke disabling dithering. One of those!

Edit:
Disable still seems to work fine here.

mojave fixed it.

[mojave]

I don't think your No Dither image is actually correct. Or I broke disabling dithering. One of those!

I tried to fix a text formatting issue and decided to upload the picture again for no dithering. I then uploaded the wrong picture and realized it after I had posted. It is fixed now.

[Awesome Donkey]

The next build will offer full-strength TPDF dithering (ie 2LSB as people like to call it), but not enabled by default. There is a new option in Audio -> Advanced -> Dither Mode

Huh, how about that? Hopefully that appeases everyone...

[mojave]

Can 32 bit be added to the bitdepth simulator? I have a 32 bit DAC. Maybe it ends up below the noise floor that can be shown in SPAN.

[Hendrik]

Can 32 bit be added to the bitdepth simulator? I have a 32 bit DAC. Maybe it ends up below the noise floor that can be shown in SPAN.

Sure, why not. I also extended it to 2-bit, which short of adding a lot of audible noise, actually gives quite good output still (with TPDF)
You can clearly hear the difference between the old mode and TPDF in 2-bit. TPDF is much better at 2-bit. Not very representative, but still a neat effect.

For 32-bit, SPAN goes to -180 dB, which shows a bit of noisefloor then. But it appears to be at its limit of precision here.

[bobkatz]

Can 32 bit be added to the bitdepth simulator? I have a 32 bit DAC. Maybe it ends up below the noise floor that can be shown in SPAN.

Can you actually send 32-bit fixed point information to the DAC through USB?

[mojave]

Can you actually send 32-bit fixed point information to the DAC through USB?

ASIO output is fixed point. Even with 24-bit DACs, most of them request 32-bit data which is why JRiver has the option in Audio Setup > Device settings > Device uses only most significant 24-bits.

I use the MOTU 1248 which has ESS Sabre32 32-bit DAC chips.

[Hendrik]

Can you actually send 32-bit fixed point information to the DAC through USB?

USB Audio 2.0 class devices can receive 32-bit over USB, yes.

[kstuart]

You'll probably also want to disable dithering if you're trying to go for that, otherwise there are no guarantees.
But who plays without any kind of DSP anyway, and if its only loudness and volume leveling.

<Raises Hand>

When I am doing any sort of listening that qualifies as "audiophile" then I am listening to an album, where the mixing engineer has already done the volume leveling (and in fact, changing volume levels of album tracks relative to each other, is actually changing the original "mix" of the album).

Conversely, when I am playing some sort of random party playlist, no one is listening close enough to possibly have the difference between dithering methods matter.

So, I assume that:

* "Disable Volume"  and
* Blue in Audio Path

means that no dither would be used ?   ( Still have not received a yes/no answer )

Some volume changes (such as those done by volume leveling or adaptive volume) will not turn off the "blue light"

But I assume that "Disable volume" disables both volume leveing and adaptive volume...

[stewart_pk]

Hmmm ... anyone?

[blgentry]

But I assume that "Disable volume" disables both volume leveing and adaptive volume...

It does not.  Disable Volume just turns the volume slider to 100% and doesn't let you adjust it.  Volume leveling and Adaptive Volume are still active with Disable Volume set.  How this affects dither, I have no idea. 

Brian.

[mwillems]

So, I assume that:

* "Disable Volume"  and
* Blue in Audio Path

means that no dither would be used ?   ( Still have not received a yes/no answer )

But I assume that "Disable volume" disables both volume leveing and adaptive volume...

Why would you assume that when you can test in a matter of seconds (like I did before answering you above) and find out whether your assumption is merited? Hint: it's not.

Disable volume, enable volume leveling, and then look at the audio path.  You'll see that the volume leveling adjustment is still happening, and the path icon will still glow blue.  

When I am doing any sort of listening that qualifies as "audiophile" then I am listening to an album, where the mixing engineer has already done the volume leveling (and in fact, changing volume levels of album tracks relative to each other, is actually changing the original "mix" of the album).

Volume leveling doesn't affect intertrack dynamics when used to play an album, it only affects the global level of the album, which the engineer has not done in any meaningful way unless you have calibrated your system to play back sound at exactly the same volume as the engineer's monitoring rig. Absolute volume level for the entire album is arbitrary unless you know the level the mixing engineer was listening at; relative intertrack levels are important, but volume leveling does not affect them when all tracks are part of the same album. 

[mwillems]

Many DACs ask for a 32-bit input in ASIO mode and then discard the lower 8 bits (dithering down to 24).  If your DAC doesn't accept 32 bit output in WASAPI (most don't), then the two modes should be the same (although depending on your settings the dither may get done in a different place).

[fitbrit]

So now we have people dithering over dithering?

[stewart_pk]

My DAC does not ask for a 32 bit input. My USB to SPDIF converter does but at the same time it only outputs 24 bit to the DAC.
The way I see it if I output from JRiver with dithering on at 32 bit the USB to SPDIF driver is also going to dither down to 24 bit upon output, sound reasonable?
But perhaps if I output at 24 bit from JRiver in a 32 bit package then the USB to SPDIF converter will simply discard the 8 bit zeros and output at 24 bit untouched, or perhaps not.
Once again I don't expect this to be audible or to an extent other than subtle.
But I would like to know what is the technically superior option as it should at least help me understand this stuff better.

[stewart_pk]

Volume leveling and Adaptive Volume are still active with Disable Volume set.

Only if they are actually switched on I presume?

[glynor]

Yes.  Of course.

[kstuart]

Okay, for "audiophile" album listening, I have "Disable Volume" and nothing checked in the DSP, except Output Format (which is "no change" except for DSD which already has noise shaping).

Volume leveling doesn't affect intertrack dynamics when used to play an album, it only affects the global level of the album,...

But of course, when I am listening to that album, I manually adjust the volume to be .... drum roll.... the volume at which I want to listen.

So, no need for volume software.   When using MC21 for random/party listening, then there is no "audiophile" aspect, and no one is paying attention closely enough for the difference between two dither formats to matter.

Thus, I think that you have to be in a particular situation to be using DSP while you are doing audiophile listening.  I'll take your word for it that you need it.

[Hendrik]

Dithering is never not performed as long as its enabled. But this is not bad. Even if you're playing 100% bitexact, then the signal ought to have baked-in dithering, or you run into the same old problems. If it does have baked in dithering, then re-dithering will just "overwrite" the original dithering. Dithering twice does not increase the amount of noise, it just overwrites the original noise.

If you want to be able to output absolutely bitexact at 100% volume, you should use the old "JRiver Bit-exact Dithering" option, which is tuned in such a way that even if its applied, it will not change a signal if its output without *any* changes (same input/output bitdepth, no volume changes or any other DSPs) - hence the Bit-exact in the name.
TPDF is a stronger dithering, and does not have this characteristic.

Dynamically turning off dithering when MC thinks the signal is untouched is rather risky, as the number of conditions for that are way too long and complex to get them 100% right.

Its also important to remember that at 24-bit output, the dithering noise is at a level somewhere south of -160dB

[jjazdk]

Thus, I think that you have to be in a particular situation to be using DSP while you are doing audiophile listening.

That is one strange assumption, though it might apply in your special case.

Audiophile listening or not, in my setup DSP is always needed, due to my active filters and EQ.

[v_erich]

Hi,
I have a small question:

When I use JRiver to convert music (e.g. convolving offline), is dithering used for the new stored e.g. WAV file?
You have to convert the 64 bit float from the internal computing to what is configured for converting, so dither should be applied there i think.

BR
Erich

[Hendrik]

When I use JRiver to convert music (e.g. convolving offline), is dithering used for the new stored e.g. WAV file?
You have to convert the 64 bit float from the internal computing to what is configured for converting, so dither should be applied there i think.

Yes, dithering is used there as well.
We only have one function that converts the 64-bit internal audio back to the output formats, and its used everywhere a conversion is needed - so they all automatically inherit the dithering.

[bobkatz]

We can't beat the laws of physics. Dither is, fundamentally noise and if you first dither a long-wordlength source to a given fixed-point result and then for some accidental reason you dither it again without changing the gain, then the noise floor will increase by 3 dB. At least with accurate TPDF randomization. If the gain has been changed before adding the new dither, then the noise floor increases by an RMS sum of the amplitude of the previous dither modified by the gain change plus the new dither.

As a mastering engineer I do a lot of manipulation (hopefully for the betterment of the sound). I've done considerable listening to cumulative 24-bit dither and it takes a few generations before any audible degradation may be noticed. However, cumulative 16-bit dither is usually frowned upon as it does degrade the depth and dimension enough to be audible (to a keen ear). It's quite benign to add 24 bit dither to previous 16-bit-dithered material and redithering is the correct way to listen to CDs with a digital volume control. Some systems automatically defeat the dither when the source is played back at exactly unity gain with no processing, but I wouldn't get hung up over that feature since properly-implemented 24 bit TPDF dither is really really really benign!

Let's refine this term "overwrite" that you used: The "cancel out" terminology some people refer to comes from this situation:  Supposing you start with a dithered 24 bit signal. You change the gain or equalize or whatever. This produces a 64-bit float output so you must dither to 24 bits again on the way to a 24-bit DAC. Your new dither is necessary and the old, previous dither becomes "irrelevant", but not, precisely "cancelled out" or "overwritten". That's a misleading term. The noise of the original dither still exists in the total result, but it no longer has any distortion-reducing effect after the new gain calculation took place.


Hope this helps,


Bob

Dithering is never not performed as long as its enabled. But this is not bad. Even if you're playing 100% bitexact, then the signal ought to have baked-in dithering, or you run into the same old problems. If it does have baked in dithering, then re-dithering will just "overwrite" the original dithering. Dithering twice does not increase the amount of noise, it just overwrites the original noise.

If you want to be able to output absolutely bitexact at 100% volume, you should use the old "JRiver Bit-exact Dithering" option, which is tuned in such a way that even if its applied, it will not change a signal if its output without *any* changes (same input/output bitdepth, no volume changes or any other DSPs) - hence the Bit-exact in the name.
TPDF is a stronger dithering, and does not have this characteristic.

Dynamically turning off dithering when MC thinks the signal is untouched is rather risky, as the number of conditions for that are way too long and complex to get them 100% right.

Its also important to remember that at 24-bit output, the dithering noise is at a level somewhere south of -160dB

[Arindelle]

have you been following this post? http://yabb.jriver.com/interact/index.php?topic=100095.0