Hi, I always leave the 64bit double processing engine on  and dithering off during export. Recently had some subtle trouble with audio quality in a final mix export, it was subtle but bit obvious in the strings section. I record and render in 24bit. Since I was not satisfied with the way the strings section sounded after the  first export I turned on the dithering and made 2  more exports with tringular and power-r2 settings. The way the strings sounded improved in the later exports. Again, it was subtle but obvious to me since I was working on the mix for quite some time. 

Previously I thought there is no need to turn on dithering during export since I was exporting  the mixdowns in the same 24bit resolution. I have read many advices on using dithering only during mastering. But recent googleing has educated me that daws by default work at 32bit resolution so there is some internal conversion doing on from 32bit to 24bit even if you set the project at 24bit. Is this true? And is true with sonar?

So this also got me wondering, if that is the case each time I am rendering using 64bit double processing engine sonar has to convert the audio from 64bit to 24bit? Am I right? This can cause even more low level distortion? and thats why my export needed dithering to get it sound right? Please someone knowledgeable enlighten me. 

No, the only downside to 64-bit processing is that it places greater strain on system resources: more RAM used, more CPU cycles, more disk storage.
 
Dither is irrelevant in either 32- or 64-bit processing. There is some debate over whether it is relevant to 24-bit integers, but a strong case can be made that it is also a non-factor in that scenario. Yes, your 64-bit files do need to be truncated to 24 bits, but the resolution is high enough that no significant precision is lost. Remember, all this is happening at ridiculously quiet levels way below the level of audibility.

Generally dithering has an effect when you are exporting to integer formats (16/24 bit).
If you intend to use it for mastering you should export to 32 or 64 bit float with dithering turned off.

I just get clicks and pops if I turn that on. So it stays off.

Here goes - I'm fast becoming the expert on this one, although some on here will howl me down - I am apparently a trained researcher these days (I've used a content analysis to get across this topic, but I probably need to do more qualitative study into the topic, scientific)...but since I taught myself and I'm in a protracted war with the university system here in Oz...who knows but here goes!

Firstly, well done on hearing there's a problem when going from 64bit FP to 24bit. This is not a condescending statement - some on this forum say you can't hear any differences because you only have 24bit converters. In fact it was on the strings or the sampled strings that I first heard the problem too.

One of things I cannot stress enough to you is switch on the 64bit fp record button, this is where you will encounter one of the problems that you can encounter down the line. I'm not sure how Sonar's internal dither system works, but if you're recording at 24 bit but processing at 64 bit...there might be dithering going on here. So not only are you dithering at the mastering stage, but also during the processing stage. I'm speculating, because I don't know your settings. Also, you've picked up on an important notion that I only realized myself a few weeks ago. This is not all audio programs programs process 64bit FP audio files. It was a revelation, when I heard this mistake for the first time. You see I use Vegas Video for my moving image editing...and I didn't realize that it's internal audio processing is done at 32bitfp. I would import my 64bit fp master into Vegas, but I kept hearing something, probably similar to what you're hearing...there was a graininess at the very outer edge of the resolution. It bothered me, but I thought it might just be a crap mix...it wasn't. About 2 months ago I decided to see what happend when I slapped Vegas Video's dither plug over the top of the master (I never put dither on my 64 bit masters before then you see), and it was a revelation. Gone was that nasty graininess you're hearing.

Here's my advice to sum up - first always record, mix and master at 64bitfp. Only dither when you need to. So if you're going to use the master file in a film clip check what bit depth the program processes audio and dither accordingly. Now I make a few mistakes here and there, it's why I get laughed at on this forum...but I'm still trying to work out whether to dither for MP3 and compressed audio formats...do some experiments and use your ears to determine whether or not to do so. At the moment I use 64bitfp master files, which I upload to soundcloud and convert to MP3 format. I will have to do some listening tests on this one. I also recommend junking powr dither - it is absolute rubbish. I will tell you why - when I slapped Vegas Video's dither over my 64bitfp master file...it was the first time ever I heard what the dither was doing. If you do a little research, you will find dither has some really basic settings...and these settings resemble film filters in many ways. Think gaussian blur, which is a setting on my dither. Being able to hear what the dither does, means you can set the dither accordingly and indeed not chop too much off. So my dither has rectangle, triangle, half triangle and gaussian off the top of my head and you can shift the noise floor to the top end and across the spectrum...off the top of my head again. It is so cool when you hear what you're doing, because you can keep more of the definition...which is super important if you're doing the amount of processing I do. Think the analogue emulation aesthetic, binaural time based effects and other audio ear goodies :). Its also why I suggest you record at 64bitfp too, it returns the digital realm to the analogue realm. You have probably more dynamic range (internally) than the best analogue system with 64bitfp audio files. And if you use Melodyne or digital varispeed - the manipulation doesn't create artifacts or a graininess again. And here's a clue too as to why you should record at 64bitfp audio, and it was my first clue in better understanding the 64bit digital audio paradigm. Are there times when you wouldn't use 64bitfp audio files? Yes! Let's say you're recording an acoustic troubadour with minimal processing...I would turn off 64bitfp and if I had good microphones and preamps use these to massage the sound. I would also switch off 64bit FP if i was going to do all external processing - going in and out of the box. You will get no benefit of the 64bit engine and audio file paradigm then and in fact it's quite possible artifacts could be introduced if you do.

Finally, to fix the issue you have I suggest this - switch on the 64bit fp audio file button, and make sure you've hit the up upsampling or x2 button and enabled the plugin to process with upsampling...upper left hand section of the plug. Most plugs apparently these days have this already punched in, but with modern computers and SSD drives these days...there's no point in not switching on these devices (unless as I said above). Then freeze each of your tracks, what this will do is wrap your 24bit audio files into 64bitaudio files...giving you the added benefits of 64bitfp audio files. This should fix the graininess you hear on your strings. Make sure you turn off Sonar's internal dither first...but maybe experiment and do an a b test. So freeze a track at 64bit fp with and without Sonar's internal dither on. I haven't really tired this, but now I've mentioned it...I will probably give it a burl.

Hope that helps:)

Ben.

"Safe" approach is to minimize format conversion, for that you need to realize which format is used by different components in the signal chain.

sonarman1
I record and render in 24bit.

An audio interface native record and play format is 24bit. From that, the last 4-8 bits are garbage (in the best case there is "good" white noise). So recording into 24bit is fine, you just record what you get. There is no magic way to improve that. And that is what defines the "Project bit depth".
 
DAW process the signal in 32 or 64bit FP format. While a DAW internally theoretically can do what it wants, VST by design support 32 and 64bit FP only. To preserve as much as possible, render into the same format. In this case you "record" (out of the DAW) what it produce. In practice, 64bit (FP) make big sense during processing only, "truncating" it during rendering to 32bit (FP) is fine. Only under extreme conditions and with cranked environment you may notice the difference. I mean when you can hear the difference, even subtle, that has other source.
Note that you should set the format not only during "Export" rendering, but also (and in fact more important) for intermediate rendering: "File/Audio Data/Render Bit Depth".
 
Rendering into 24bit can be problematic. But you still should not hear any difference if you do that under normal conditions. Check the processing chain from the source to the place where you notice the problem. Probably you will find the evil (bad VST, intermediate rendering to 24bit -40dB level signal, etc.).
 
If you do not trust me, at least trust Noel. Do not try to solve problems with dithering. It is good for final conversion into 16bit and even there should produce subtle effect and for demanding material only.

Wow, this stuff is so far above my head. What about music for full orchestra? Should dithering be used there? I usually just go with the default setting. Should I change it?

michael diemer
Wow, this stuff is so far above my head. What about music for full orchestra? Should dithering be used there? I usually just go with the default setting. Should I change it?

No. The violins will supply all the dithering noise you need.

bitflipper

michael diemer
Wow, this stuff is so far above my head. What about music for full orchestra? Should dithering be used there? I usually just go with the default setting. Should I change it?

No. The violins will supply all the dithering noise you need.

Yeah, you got that right! The low strings also do a great job muddying up the mids, too. Strings remain the most challenging thing to get right. but then I have heard actual performances where the high violins sound terrible. Just too many instruments, no way they can be in tune and in time with each other.
 
So, I guess I should turn dithering off altogether? 

These discussions on dithering and hearing the difference are interesting. For dithering meant to apply to export, there is an easily repeatable experiment that can be done. The OP would be in a very good position to try this.
 
Export multiple files, each time using a different dither setting. Twice for example: once with no dithering and once with a dithering option that sounds different. Then in a bare project, put those two files, one out of phase. Based on my understanding of the definition of dither, those files should cancel out down to the least significant bit.
 
The resulting file should be noise, and it should be very hard to hear when played back at the same volume that would normally be used to listen to the exported original. Certainly if the original file does not use the dynamic range available in the original format (e.g. 24 bits), the dither becomes a larger proportion of the signal, and may cause audible artifacts.
 
In any case, if samples in the resulting file are larger than '1' (literally the value 1 in a possible range of -838608 to +8388607) there could be a bug in the dither code or an error in the application of it. (Edit: there are some controls on the timeline to zoom vertically on the waveform to see 1's and easily see samples out of that range)
 
Does the canceled output sound like noise or does it have any pitch or audible pattern (turn up the volume to hear it)? The noise can be subjected to frequency analysis to see if the spectrum looks like noise in the frequency range predicted by the description of the dither option.