Skip to main content

kinda wondering...

what happens when one dithers a (legal) 32-Bit FP file all the way down to 16-Bit w/ their favorite dither box ?

reason I ask is...I'd been used to (like everyone else, I suppose) bouncing out of the DAW (SONAR) at 32-Bit FP...then brick walling to legal in the mastering rig (WaveLab) while rendering to 24-Bit at the same time...then, dithering (UV22HR) & rendering to 16-Bit ftom 24.

[i had been doing this 'cause I assumed (& had heard) that A) the quantization error from 32 > 24 is miniscule & well below the audible range & B) it's a legal file after limiting anyway so I may as well render to 24...why waste bits.]

I did one today whereby I kept the file size/format at 32-Bit FP during the brick wall phase...(so that it was also 32-Bit FP after limiting)...& then dithered all the way down to 16 from 32 as the final stage. It sounded audibly way better (in comparing both methods) to the old method (24 > 16).

most notable was a more accurate reproduction of transients (snare, cymbal crashes, et al.) [i had used elephant 2 on clip for the limiting phase].

guess I'll keep this new (new to me, that is) method...but I'm wondering what's goin on technically w/ it ?

thanks,

mark4man
MoonMix Studios
ADK Pro Audio Core 2 | Intel DP35DP MoBo/Chipset | Intel Q6600 Quad Core CPU | 4 GB SuperTalent DDR2-800 CL5 RAM | Seagate 160 GB SATA II Primary HD | Western Digital 320 GB SATA II Audio HD | Lynx Aurora8 ~ AES16 | Universal Audio UAD-1 | SONAR PE 7.0.2 | Cubase LE | WaveLab 5.0.1b | NI Komplete5

Topic Tags

Comments

Cucco Mon, 07/06/2009 - 10:43

Well, thanks for answering my question.
Seriously - what in god's name do you mean by legal file?
Perhaps if you wrote a coherent post, you'd get meaningful answers.
You use the term "legal" a number of times in your post and it's a term that I've never encountered after 20 years in the business.
Now, how about you answer my question and I'll try to decipher the rest of your post.

Cucco Mon, 07/06/2009 - 10:54

And yes, based on what I could discern from your first post, a single stage of dithering is always preferred to multiple stages of dither.

When you record at 24 bit, you're mixing effects and effecting volume envelopes at a 32bit float level anyway (thanks to the 32 bit processing native to your OS, processor and DAW program.) Even if you "record" at 32 bit, you're actually recording at 24 bit and appending 8 '0's at the end of your word. This is due to the limitations of your AD converter. Most AD converters operate at 24 bit (many technically less. Exceptions would be DXD or 1 Bit (DSD or Delta Sigma) converters) and output a 24 bit word.

When you mix in the DAW, those 8 slots are filled up by valid information. The DAW, if told to render the file to 24 bit, will dither the signal. Then, if you render again, you are adding another layer of dither. This is two separate noise prints laid on the same file at different amplitudes. If you're using the same noise shaping profile, it's likely you're hearing the effects of some bunching up of some noise in key frequencies.

Short version of the story:
Dither once.

Now, seriously - what do you mean by "legal" and "legal file?"