ORIGINAL: Mr. Ease

Just as a side note, as it has been mentioned hereabouts. I still cannot believe that dither is necessary when converting to 24 bit from either 32 or 64 bit floats. This is purely because the dither (in the 2 LSB's) is still way below the inherent noise of any 24 bit D-A converter. 1 LSB of truncation OR rounding noise is around -140 dBFS and insignificant to even the best converters.

Bingo. The noise floor of any electronic circuit is way higher than that.

As a followup before I go to classes...

I seem to recall our objective is to determine whether to apply dither to 16bit already mastered source files when reauthoring a Red Book CD. Specifically gain and attenuation processing would place these files in float at a 24 or 32 bit resolution internal to the host. And our question seems to focus on whether the internal host's application of dither during the gain or attenuation in the engine would return audio that required additional dithering in the render to a 16bit output file.

From what I recall one of the major points centers around a 13db gain or attenuation. I believe one of the sticking points for me is to consider any mastered audio that would need more than 5db of gain or attenuation was not mastered in my opinion - 13db is practically 200% increase or decrease in energy. I'm not stuck on this point entirely other than to refer to what would clearly be an improperly mastered track to start with, and if there were 13db of gain needed in the new master- the narrow distance between the original dither and the gain in such a file would be frightening.

However to Richard's well considered point - let's not change the focus from art to science here, and it is essential to know what is proper given the files as described without consideration to 'how' they got in that condition.

I'll say that I still feel that I would not apply dither, but at the same time I understand Richard's angle on the process. I'm feeling like this is the most artistic scientific conversation I've had in a while. I'll let y'all know if and when Bob returns my email.

-D

Richard et al;

Bob sent me a message this morning, and as I predicted on page 1 or 2 of this thread...I am wrong: Here is his reply!

Don;

I am sorry to tell you that you are wrong and the poster from the UK is right. It's all right there in the dither chapter! Wordlength ALWAYS grows if you make a change or a calculation. All that wonderful "previously dithered" material loses it if you make a change to it. So you MUST apply dither at the end of any series of calculations when you reduce the wordlength back to 16 bits, regardless of the original source wordlength of the object. It's not nice to fool mother nature :-(. Compilations are particularly problematic in this regard, and when making the tiniest change, perhaps it might be better not to make any as the tradeoff (added veil to the sound) of adding another layer of 16 bit dither might make it not worth the trouble to make the change.


Hope this helps,


Bob
--------------------------------------------------------------

Bob's last comment about the unique problems associated with compiliations really captures my original concerns and his comment about it may perhaps be better to make no change at all.... identifies my original basis for not wanting to use dither. But as he says you can't full physics and any change will result in the need for a dither algorithm.

Thanks Richard and Bob for their attention to accuracy.


-D

Nuts! I said that Don was right. I can't say about what even when I said it I was looking more at the percentages. Now Don is in my book still right it was me that was wrong. Don't ask me what I was wrong about because I have no clue. Don I have been wrong often so being use to it make it my error.

If this were me doing this, I would:

Handle the formats seperately.

Start a Sonar project in 16-bit, Import all the 16-bit files in, process as necessary, and Export to a single 16-bit file, but DO NOT DITHER.

Start a Sonar project in 24-bit, Import all the 24-bit files in, process as necessary, and Export to a single 16-bit file applying an approriate DITHER.

Then bring both 16-bit Exported files into an appropriate Mastering Audio Editor (such as Soundforge) and segment all the tracks, etc OR send to a Mastering House for completion.