Topic: Dither when adding gain to a 16 bit file?

[Sebastian]

Hey,

I have been thinking about this for a while and I don't know what the best thing to do is...

I am currently transferring some old DAT tapes. I'm recording them using WaveLab to 16 bit wave files. But sometimes, I want to add some gain and/or EQ. The question is: Should I dither the final mix back to 16 bits after I've finished all my signal processing? I once heard that all signal processing is done at higher resolutions, so dithering the results sounds logical to me, even when starting out with a 16 bit file.

Any advice is greatly appreciated. I'm pretty much lost...

[live2496]

Some may disagree, but after giving this some thought, I don't think you need to.

Here's my reasoning... If you increase the level, the noise level goes up as well. Whether that be noise from the circuit or the original dither from the converter (all converters add dither when sampling). Any newly added dither should be at a level lower than the existing noise if you amplify the recording in post processing.

That said, I don't think it would hurt anything to redither either. It would be at a level that you cannot hear. The original recording's noise should mask it anyway. Again, assuming that you boosted it.

[Gordon]

maybe I'm missing something but if you are recording (dat transfer) at 16 how can "dither back to 16"?  you would have to "up dither' first.   if it's a digital transfer (it should be) then is not advised.   are you really talking about sample rate?  if so edit at 48 (if that is the original) then resample to 44..

[F.O.Bean]

I think hes talking about programs like WL/SF/etc that make TEMP FILES at 32-bit float and 24-bit TEMP FILES. Do you mean do you have to dither the file back to 16-bit after doing any editing at higher resolutions other than 16-bit like at 24-bit or 32-bit float

If so, then the answer is NO, you DO NOT have to dither back to 16-bit. That is STRICTLY for the temp file and you'll have a 16-bit file after all of the processing is done, so youre good to go brotha

[JasonSobel]

I disagree with the responses above.  If you transfer a DAT, you've got a 16 bit file.  If you then do some EQ or other post-processing, depending on your software, that processing could be done at 24 bit, 32 bit, or even 64 bit resolution.  after ALL the processing is done, you'll have to get it back to 16 bit, and dithering is the best way to avoid small quantization noises from just truncating that signal.  now, some software may do that automatically for you, and other software will allow you to choose what dither to use (or, even allow you to choose no dither and just truncate, which I wouldn't recommend, but apparently some other folks would).

[Sebastian]

Thanks for all your responses! +Ts all around.

I think Jason was right. I just did a quick test with a sine wave file using Wave Lab's bit depth meter.

When playing the unedited 16-bit wave file, the meter shows this:


However, when I change the gain a little bit (-0.1 dB), the meter changes to this:


I think this clearly shows that I should really dither all post-processed signals. I think I'll just apply some small amounts of dither from now on.

[Scooter]

def dither at the end of the chain as stated.  in wavelab, its doing any processing at 32bits, so you gotta dither back down.  the bit meter confirms!

[F.O.Bean]

def dither at the end of the chain as stated.  in wavelab, its doing any processing at 32bits, so you gotta dither back down.  the bit meter confirms!

I dont know about that fellas. I know the bit-meter showed 16bits vs. 24bits, but it kind of doesnt make sense like that. I have opened 16-bit files in WL 4.0/5.0 MANY TIMES in the past years when I was still xferring DAT tapes and when I recorded to my trusty old Nomad JB3. And I NEVER dithered back down to 16bit when I was doing editing at 24bit or 32-bit float. Thats why its called a TEMP FILE, because it just upsamples/dithers to 24bit or 32bit, and does the processing at the higher resolutions, and then since its a TEMP FILE, once the processing is done at 24bit or 32bit float, then oncer the editing is done at the higher resolutions, the file is still a 16bit file. All you did was take the original 16bit file, process it(change gain/resample/etc) at 24bit or 32bit float, and then once the processing is done at the higher resolutions, you are left w/ a 16bit file once again. The 24bit and 32bit float are JUST FOR PROCESSING/EDITING, and the way I ALWAYS understood it, is that I would NEVER have to dither back down to 16bit after the processing/editing is done.

So, have I been doing it 100% WRONG the past 5-6 years Have I really been truncating the extra bits

It just doesnt make sense to me to dither a file back to 16bit, when it was 16bit ORIGINALLY. ALl tahtw as done was processing at higher resolutions so the editing/processing is as clean as possible and no "artifacts" are introduced.

Hell, since I started recording at 24-bit back in June 2006(2+ years ago), I have only been doing my TEMP FILES at 24-Bit instead of 32bit float since when dealing w/ the 2GB filesize limit, 32bit float eats up  WAY TOO MUCH space and near the end of the file that is 2GB, WL 5.01b stops the processing and states that there isnt enough space on the HDD, but its really because when using the 32bit float, you hit the 2GB filesize limit before processing is done since 32bit has ALOT more info/data/resolution vs. 24bit TEMP FILES. Since I started using 24bit TEMP FILES instead of the 32bit float TEMP FILES, I havent had not ONE ISSUE regarding hitting the 2GB filesize limit while processing/editing

So if yinz are recording at 24-bit and are normally recording a couple of hours and the 2GB limit is normally reached when you record, then Id HIGHLY SUGGEST switching your TEMP FILES from 32bit float to 24bit, and you will thank me later That way you wont hit the 2GB filesize limit when processing/editing your recordings NOTHING was worse than when my set/show/recording/whatever would be ALMOST complete in WL 5.01b when I was processing/editing the recording, and then when it would be about 95% done, WL 5.0 would hit the 2GB filesize limit becuase I was using 32bit float tempo files. So then youd have to switch to 24bit temp files and start the processing/editing ALLL OVER AGAIn. So instead of worrying if my recording would hit the 2GB limit when ptrocessing, I just permanently kept my temp files at 24bit. Does that make sense everyone

[JasonSobel]

Bean - as soon as you start processing the files at higher bit resolutions, you need to somehow bring the files back to 16 bit.  They're called TEMP files for processing, simly because the whole idea is to bring it back to 16 bit when you're done.

So, have I been doing it 100% WRONG the past 5-6 years Have I really been truncating the extra bits

it depends on how your software is setup.  Some software will do all processing at 24 bits (or 32 bits or whatever) and then truncate back to 16 bit after every step.  that's bad.  other software will dither back to 16 bit after every step.  that's also bad.  you want to make sure all the processing is done at the highest bit resolution possible, and then have dither applied just once, as the last step.

although, its good to have control over it.  when I transfer a DAT, I've got a 16 bit file.  usually, the only processing that I'll do is add fades at the start and the end.  Now, I don't want to apply dither to the whole file, because only the few seconds of fade in/out is at 32 bit (the rest of the file, unprocessed, is still at 16 bit).  So I turn off the dither, and accept as a compromise that the few seconds of fade in/out will be truncated back to 16 bit.

So if yinz are recording at 24-bit and are normally recording a couple of hours and the 2GB limit is normally reached when you record, then Id HIGHLY SUGGEST switching your TEMP FILES from 32bit float to 24bit, and you will thank me later That way you wont hit the 2GB filesize limit when processing/editing your recordings NOTHING was worse than when my set/show/recording/whatever would be ALMOST complete in WL 5.01b when I was processing/editing the recording, and then when it would be about 95% done, WL 5.0 would hit the 2GB filesize limit becuase I was using 32bit float tempo files. So then youd have to switch to 24bit temp files and start the processing/editing ALLL OVER AGAIn. So instead of worrying if my recording would hit the 2GB limit when ptrocessing, I just permanently kept my temp files at 24bit. Does that make sense everyone

instead of comprising processing bit depth (choosing 24 bit processing instead of 32 bit), you'd be better off using software that doesn't choke on temp files large than 2GB.  that issue is fixed in WaveLab 6.  or switch to something completely different.  I made the switch from WaveLab to Samplitude about 6 months ago and couldn't be happier

[DSatz]

Jason Sobel's reply is correct. The best way to process a 16-bit recording is to use software which allows more than 16 bits of precision for the result. Then when all such processing has been done, the result should be brought back to 16-bit precision with dither.

Keep in mind that any processing that you perform on a digital audio recording, other than muting or DC shift, etc., will create samples with greater than 16-bit precision as its intermediate result. This is analogous to what happens in decimal arithmetic when you multiply 0.5 by 0.5, for example; the number of significant digits in the result (0.25) is generally greater than the number of significant digits in either operand.

So if you're EQing a 16-bit recording and the filter coefficients have 16 bits of precision, each sample of the result could have up to 32 significant bits. That result then needs to be dithered--not just chopped--back down to 16.

--best regards

[rowjimmytour]

maybe I'm missing something but if you are recording (dat transfer) at 16 how can "dither back to 16"?  you would have to "up dither' first.   if it's a digital transfer (it should be) then is not advised.   are you really talking about sample rate?  if so edit at 48 (if that is the original) then resample to 44..

I believe this post is correct and sample rate is what they meant and not dither since they are transferring DAT most like from 16/48 to 16/44.1 for CDR.

[Gordon]

Thanks for all your responses! +Ts all around.

I think Jason was right. I just did a quick test with a sine wave file using Wave Lab's bit depth meter.

When playing the unedited 16-bit wave file, the meter shows this:


However, when I change the gain a little bit (-0.1 dB), the meter changes to this:


I think this clearly shows that I should really dither all post-processed signals. I think I'll just apply some small amounts of dither from now on.

I just did the same thing.  I had no idea it did this.  the file still shows as 16 bit so I just assumed it was temp only.  one thing I didn't like about audition is that you had to dither so much.  never knew I needed to with wavelab at 16 bit.  learn something new every day.  wavelab obviously does a good job going back to 16 cause I bet 95% of us never knew this or dithered when working with 16 bit files.