Dithering is detrimental:
This line from your post only supports my statements:
"Dither, contrary to what has been said before, is not detrimental to audio quality. Lack of dither can be severely detrimental."
It is clear the process of dithering is trying to correct issues while processing the signal. Saying adding more "empty filler" to what is already "empty filler"does not there makes it better.
The only argument is subjective ... if you like the "distortion" it add because you perceive it as better, that is harder to argue.
However, if you listen to the original recoded source ... and it sounds "different" ... then you have adulterated the signal and added distortion. With digital it is a bad thing. I hear thinner, grittier, less organic, unnatural sound when I listen to overly processed digital signals.
SACD encoding is another culprit. To me, it sounds awful ... not natural or organic and with hellacious artifacts due to the ultra high sampling rate. However, even many audiophiles seemed to like it because it sounded more "open with more detail & imaging" .... but at what cost. Loss of organic sound quality and a grit and hash that makes it unlistenable to me.
I can get better imaging, detail, and resolution with much better organic sound quality from a properly recorded CD. Again, just listen to any Mapleshade Audio or Wildchild CD. Most sound better than any SACD Discs I have heard.
The only time I actually hear this work to any level of satisfaction is with tubes. My tube equipments sounds great ... a pleasing form of analog distortion ... yet organic and true to life sound ... not processed in a harsh stair-stepping PCM or Lossy Compression sort of way.
I believe some of it has to do with Accounting. I am at odds with accounting and many cases marketing in most companies. Saving $2.00 on a DSP Chip in a $4,000 HDTV and getting poor results because of it is one reason why engineering sometime choose to lay down in defeat.
Believe it or not, I have spoken to engineers that work for big companies who actually want to do the right thing. However, they are overruled by Accounting who wants a cheaper DSP and Marketing who wants the HDTV Panel to look more "red" because it will look & sell better on a retail floor.
That is why I own so many Pioneer Elite and Kuro 9G Monitors. Nothing made to date can beat them.
Such is life..
Originally Posted by AndreYew
Roger already addressed this with addition, but this is well-known to computer programmers. Every time you add two numbers, you need one more bit of precision, because you could potentially double the maximum value of the numbers being added. And every time you multiply, you could need twice as many bits to hold the result.
The Motorola DSP56001 series of DSP chips, available back in the late 80s, had a 56-bit fixed-point accumulator for this reason. State of the art audio DSP today uses double-precision floating point with 52 bits of mantissa and 11 bits of exponent.
But your point is taken, because there are some very complicated DSP processes (like high quality EQ or filtering, especially at high sample rates) that require many, many more bits (like hundreds or thousands) to perform accurately because there are so many computation steps, and the people who implement them do have to do the proper math in order to get those DSP routines working.
And before the results are truncated down to 24 bits or whatever the delivery media will hold, dither is added. Dither, contrary to what has been said before, is not detrimental to audio quality. Lack of dither can be severely detrimental.
As for upsampling's accuracy, using a very high quality upsampler (Weiss Saracon), people have obtained bitwise null results going from 96 kHz to 44.1 kHz back up to 96 kHz, using linear phase filters. The files did not null when using minimum phase filters (like the ones Dolby's trying to push). Don't like accuracy? Use minimum phase, no-ringing filters.
A null test means subtracting one file from another and seeing what remains. The test above had zero difference in the files after one file had been through a downsampling and an upsampling. Here is a link to the claims by Bob Katz, someone with not inconsiderable audiophile cred:http://bach.pgm.com/pipermail/proaud...ay/015334.html
And yes, you have to be very careful when performing null tests, lining both files up exactly, which can sometimes be impossible if the DSP introduces sub-sample delay: http://bach.pgm.com/pipermail/proaud...ay/015347.html