[FFBookman]

IMHO Dither is the prime suspect for changing the sound when downsampling.
 
I record at 2400k or higher bitrate, sometimes as high as 5800k for stereo PCM.  I mix very closely and pay lots of attention to delays, reverbs, room, air, placement, etc.. 
 
Then I down-sample and apply dither. I have probably 4 different algorithms and each of them change the sound, especially the "size" of the sound -- the depth, width, clarity, and resolution of the various voices and instruments.
 
Since my rig is for production and has decent but not great converters (focusrite) I then send the full resolution versions out to my mastering engineer. He does a similar process on his much more expensive rig and delivers to me the file labelled A/B/C with various dithers applied.
 
There are huge differences in the quality of the dither algorithm. Dither is fuzz used to cover the downsampling errors. Well not errors. Going from 24/192 to 16/44 is trying to squeeze 16 million datapoints into 64 thousand slots. A lot is lost.
 
I make music of my own, I also record and mix music created by others. I've been doing this on and off for about 25 years now. I learned by cutting tape (1" and 1/2", never cut 2") and was running Pro Tools v1 or 2 I think. I have plenty of experience with the formats and the conversions.
 
This little group of people who don't believe anyone can hear beyond 1400k bitrate is ridiculous and fascinating to me. Even digital recording systems in the 1980's did 20bit and over 44k sampling rates.

 

[arnyk]

  IMHO Dither is the prime suspect for changing the sound when downsampling.
 
I record at 2400k or higher bitrate, sometimes as high as 5800k for stereo PCM.  I mix very closely and pay lots of attention to delays, reverbs, room, air, placement, etc.. 
 
Then I down-sample and apply dither. I have probably 4 different algorithms and each of them change the sound, especially the "size" of the sound -- the depth, width, clarity, and resolution of the various voices and instruments.

 
Please detail the means of technical evaluation that led to these conclusions.
 
If they were typical casual sighted audiophile evaluations, then something better will be required to be convincing to many.
 
I really don't know what to make out of the rest of your post because of the lack of supporting detail, including which of the many possible dithering algorithms were used. Some are good, some are bad. Perhaps your choice of dither is similar to your choice of listening test methodology. 

 

[StanD]

The prime suspects are those using voodoo to explain what they can't possibly hear. Just more spurious conjecture.

 

[arnyk]

  The prime suspects are those using voodoo to explain what they can't possibly hear. Just more spurious conjecture.

 
 
The issue that people who use reliable listening tests worry about is the transition band on the low pass filter and artifacts associated with it.
 
Trouble is - the reliable listening tests say not so much...
 
Dither as an issue was pretty well spiked by some AES papers a decade or more back.
 
But with sighted evaluations, all things are possible...
 
BTW, studying this issue is one of the easiest DBTs to run. So, no excuses.

 

[castleofargh]

it's easy Arny, he rejected ABX several times already. not a chance for him to do an actual controlled test. because to judge of things changing below -90db just press play and it should be obvious... I mean you have bad ears and a bad sound system not to notice it without a test!!!!

 
and I'm sure his sound system has nothing making noise, distortions, crosstalk, or channel imbalance that would change the "size" of the room much more than the worst dither ever. I wish I had his speakers.
 
 
 
 
for normal human beings, just play a song at normal loudness in foobar, then lower the volume level abruptly by about 60db, just to feel what it sounds like down there. now go -90db enjoy all the amazing differences bookman is talking about before going back up to the original loudness. should give an idea of how unimportant all that really is.
and to make things funnier, remember it's uncorrelated noise once dithered, so pretty much the kind of sound less susceptible to be noticed.

 

[StanD]

   
 
The issue that people who use reliable listening tests worry about is the transition band on the low pass filter and artifacts associated with it.
 
Trouble is - the reliable listening tests say not so much...
 
Dither as an issue was pretty well spiked by some AES papers a decade or more back.
 
But with sighted evaluations, all things are possible...
 
BTW, studying this issue is one of the easiest DBTs to run. So, no excuses.

And what is the bit depth of dithering and how does that fit into the DR? Adding less than 1 bit of noise out of 16 isn't going to keep me up late at night. I'd bet that the OP stays up late at night pondering this.

 

[arnyk]

  And what is the bit depth of dithering and how does that fit into the DR? Adding less than 1 bit of noise out of 16 isn't going to keep me up late at night. I'd bet that the OP stays up late at night pondering this.

 
Most concerns and discussion seem to be related to downsampling 24 bit data to 16 bits.
 
The book solution is TPDF dither with perceptually favorable spectral shaping.
 
I'd be curious as to more details about this particular concern.

 

[bighat]

the story almost don't add up.

Since my rig is for production and has decent but not great converters (focusrite) I then send the full resolution versions out to my mastering engineer. He does a similar process on his much more expensive rig and delivers to me the file labelled A/B/C with various dithers applied.

you record with focusrite and then you send files to mastering engineer to make dither itb? because he have got better otb converters? lol

There are huge differences in the quality of the dither algorithm. Dither is fuzz used to cover the downsampling errors. Well not errors. Going from 24/192 to 16/44 is trying to squeeze 16 million datapoints into 64 thousand slots. A lot is lost.

dither is not exactly a covering of quantization errors, the fuzz can expand sound to noise ratio in frequency domain,
going from 24/192 to 16/44 could be like applying a low pass filter at 22khz and cuting the lowest 48db from 144db of sound.....
it should be unnoticeable.
 
 I must add that Chesky records kick ass

 

[dprimary]

I don't understand your workflow. Your data rates do not match any common sample rate/bit depth rates. Why do you down-sample and dither before sending to the mastering engineer? What are you using to down convert and dither? In the early 90's this was problem, it does not seem to be a problem anymore.
I am not aware of any production recorder in the 80's that could do 20 bit or greater recordings. The Sony, Studer, Mitsubishi, NED were 16 bit in the 80's.

 

[StanD]

  I don't understand your workflow. Your data rates do not match any common sample rate/bit depth rates. Why do you down-sample and dither before sending to the mastering engineer? What are you using to down convert and dither? In the early 90's this was problem, it does not seem to be a problem anymore.
I am not aware of any production recorder in the 80's that could do 20 bit or greater recordings. The Sony, Studer, Mitsubishi, NED were 16 bit in the 80's.

He was dreaming.

 

[FFBookman]

  I don't understand your workflow. Your data rates do not match any common sample rate/bit depth rates. Why do you down-sample and dither before sending to the mastering engineer? What are you using to down convert and dither? In the early 90's this was problem, it does not seem to be a problem anymore.
I am not aware of any production recorder in the 80's that could do 20 bit or greater recordings. The Sony, Studer, Mitsubishi, NED were 16 bit in the 80's.

I don't downsample and dither before sending to mastering engineer, not sure where you got that from.
 
I will occasionally downsample and dither w/o the mastering engineer to hear different dithering and downsampling on my own rig. But it's not high end converters like the mastering engineer has.
 
So I deliver in the highest native resolution I have and he downsamples and dithers. Occasionally he returns A,B,C versions of the file was created using different dithers, and occasionally he returns MP3 and AAC in that same 'almost blind' naming scheme.  I appreciate this stuff because I am a producer and it helps my ears.  I close my eyes and play them on the same rig that mixed them, and occasionally check them on other systems, then make my pick. Each of them sounds different. Each dither applies it's own character to the sound.
 
What part of that is so confusing?  I deliver 24/88 or 24/96, he returns 16/44 WAV, MP3, and AAC.

 

[FFBookman]

  And what is the bit depth of dithering and how does that fit into the DR? Adding less than 1 bit of noise out of 16 isn't going to keep me up late at night. I'd bet that the OP stays up late at night pondering this.

linear thinking. none of this keeps me up at night pondering. i'm mixing, playing, tracking, whatever. sleeping fine, thanks.
 
to me it's in the detail, accuracy, and depth of the material in the soundstage.  it never shows on a waveform or scope. it is so tied into the art of stereo recording, playing instruments into microphones, and mixing artificial sounds into the soundstage that you can't clearly measure it. it lives in decays and pre-delays or reverbs, attack and decay of percussion, timbre of stringed instruments (including the human voice), and the overall dynamic range to pick up breaths, coughs, studio sounds, etc.
 
higher resolutions simply resolve more material, all other things being equal. there is more there there, even if no one ever is able to reliably measure it.
 
i am one that knows a good 16bit recording sounds better than a bad 24bit recording. but given the same material, 16/44 chops off just enough to really hear it.
 
the jump from 16/44 to 24/44 is as noticeable as the jump from 24/44 to 24/96 in my ears. the jump to 24/192 really depends on the remastering, i've heard amazing and i've heard stuff that doesn't have much more than 24/96.

 

[StanD]

 
linear thinking. none of this keeps me up at night pondering. i'm mixing, playing, tracking, whatever. sleeping fine, thanks.
 
to me it's in the detail, accuracy, and depth of the material in the soundstage.  it never shows on a waveform or scope. it is so tied into the art of stereo recording, playing instruments into microphones, and mixing artificial sounds into the soundstage that you can't clearly measure it. it lives in decays and pre-delays or reverbs, attack and decay of percussion, timbre of stringed instruments (including the human voice), and the overall dynamic range to pick up breaths, coughs, studio sounds, etc.
 
higher resolutions simply resolve more material, all other things being equal. there is more there there, even if no one ever is able to reliably measure it.
 
i am one that knows a good 16bit recording sounds better than a bad 24bit recording. but given the same material, 16/44 chops off just enough to really hear it.
 
the jump from 16/44 to 24/44 is as noticeable as the jump from 24/44 to 24/96 in my ears. the jump to 24/192 really depends on the remastering, i've heard amazing and i've heard stuff that doesn't have much more than 24/96.

You are simply hearing things as an example of expectation. You are not from Krypton.

 

[arnyk]

 
to me it's in the detail, accuracy, and depth of the material in the soundstage.  it never shows on a waveform or scope.

 
How do you know that soundstage never shows up on a waveform or a scope?
 
That's quite a global claim. 
 
Have you ever looked for soundstage with a scope?
 
Do you even have a scope?  If so, which make and model?

 

[FFBookman]

Looking for soundstage with a scope --- no, I don't think I've ever tried such a thing. 
 
I don't currently own any scopes myself, just have used a few over the years in other people's rigs. I used to work at a facility with 3 TV and 4 radio stations and hung around the engineers. I was in programming and tech staff but not a radio/TV engineer myself. 
 
I wouldn't even know where to start to measure soundstage, measure delay and decays in recorded music, and to measure accuracy of signal of recorded music.  
 
My whole point is there's no quantifiable measurement for some (much) of the character, quality, and simulated size of a mixed piece of music.
 
You can spend your life measuring test tones, square waves, and pulses or you can spend it making music with with violins, guitars, drums, and synths. Or you can spend it mixing music made by others. Either way time is finite, and no, I have not developed measurements for all of the things you have no measurements for. 
 
I also am not chasing the unachievable. Looking for time-based, stereo-separation-based measurements to justify or disprove what I clearly hear (and I believe all of you do to) is pointless to me.
 
Here's more on what I'm talking about:  http://www.stereophile.com/content/2011-richard-c-heyser-memorial-lecture-where-did-negative-frequencies-go-measuring-sound-qua
 
 
So back to topic -- who here has tested various dithering algorithms when going from 24bit to 16bit with recorded and mixed music?   If you've done it, was it music you've mixed yourself?
 
If you haven't done this, just say so.  Perhaps it's where I get all of my kooky opinions from.