What kind of dithering/noise shaping do you use?

Which is what makes 24/88.2 on the Alesis Masterlink so interesting - easy downconversion.

 

Yes, they are.

I do not know if there's consensus, but dithering does help. It's just a matter of choosing the right type for the job at hand.

 

I can do 24/88.2 on my DAW...and i've got the 24/96 conversion down to where the sound doesn't suffer.

 

Well, there is also a technical explanation. This is hard to understand and it has to with statistics. Nevertheless, I can give it a try.

There is a lot fuzz about noise shaping and everybody is free to use whatever he or she wants. But the mathematics show that noise shaping is not ideal, it can produce distortion and/or noise modulation. So, I never use noise shaping.

There are some good articles about dither written by Stanley Lipshitz et al and in brief they boil down to this. Consider a multi-bit quantizer with in and output. The difference between the input and output will be subject of investigation. This difference signal can be seen as a stochastic process with the well known moments: the mean, the variance etc etc. It was found that by adding white noise with a variance of 1 bit and a rectangular pdf, the first moment (the mean) become zero. It is said that there is no harmonic distortion due to the quantization process. By convoluting the white noise with rectangular shaped pdf with itself, the second moment disappears. The variance is zero and this means no noise modulation which is correlated to the input signal. This is said to be sufficient for audio. For video there are reasons to believe that there is jet another convolution needed to make the third moment zero.

So, for me in Audition: 1 bit dithering (I am told that this means 2 bit peak-peak), triangular pdf and no noise shaping. But that is a pure technical approach.

Regards,
Jacco

 

Thanks Jacco, thats some great info

 

I see two ways to look at this.

If I rip a CD and then do some "personal remastering" at 32 bits, the original dither should still be there, thereby eliminating the need for a second dither.

or

The 8 least significant bits is affected by the processing at 32 bits, thereby risking their loss by truncation.

Anybody able to give a definitive answer on this?

 

But, I don't care about statistics. I just go by what I hear, and noise shaping to me is much preferable to just dither. I don't like the sound of 1-bit with a triangular p.d.f., and no noise shaping. It makes the music sound too veiled.

 

I've read convincing answers to support either contention.

My ears have supported the latter, applying judicious dithering on the way down again. YMMV

 

err how do I go from 32-bit to 24-bit in AA? I've never seen the option.


~0~0~0~

OK, I guess I can just type 24 on the box..

 

My ears have heard great results on not dithering twice. This, of course, depends on how much dither you use, and how it's distributed. What I have been doing, if I must go back to a higher bit-depth, is process once with dither, then go back to 16-bit wihout dither. This perhaps cancells out the dither used in that extra processing.

 

You could. I've never done it that way. I always just click "Save As" in the File menu. The problem with doing it your way is that there are like three or four types of 24-bit file types. By doing it my way, you can specify with certainty that you will pick the right one.

 

No, cosmosis, use the "save as" function und choose "options".
Just typing 24 in the Convert Sample Type box doesn't work. You still get a 32 bit file.

Edit: Grant was faster than me.

 

And, even when you do get a true 24-bit file, Audition (don't remember about Cool Edit anymore-it's been so long since i've used the older versions) still reports the file as 32-bit.

BTW, for absolute compatability, make sure you use the one that says 24-bit packed Int (Type 1, 24-bit). That is the universal 24-bit standard. It is a fixed integer format.

 

I think I'm starting to get some grip on this dithering stuff - I hope.

I've just hooked up my Technics SL BD20D to my Rotel RQ 970BX pre-stage, then into my Asus Sabertooth Z87. I'm recording in Sound Forge 9.0 in 24 96, basically shifting boxes of 45s onto my computer for storing as WAVs. I'm keeping the 24 96 masters, just out of good backup practices, but after decrackling and some very minor EQing, the tracks will be downsampled to 16 44.

Like the original poster of this thread, I'm presented with the different dithering options; half triangler, rectangler, triangler and etc, and with Noise Shaping, I'm offered the choice of high pass contour and equal loudness contour.

I've read most of this article, but it did get rather involved technically speaking.
http://theproaudiofiles.com/dither/

Which had then lead me to this article, which made things more clearer on the Noise shaping issue. Upon reading this article, it seemed I was sold on the idea of noise shaping. http://www.audiocheck.net/audiotests_dithering.php

I've also read some of this Sound on Sound article, but again it started getting too involved technically which spun my head out..
http://www.soundonsound.com/sos/may07/articles/dpworkshop_0507.htm

So after a hell of a lot of information overload, I'm coming to the conclusion that triangler will be the best dithering, and some sort of noise shaping will be necessary for smoothing the quantisation process

I'm not sure on choosing either High Pass Contour (which I'm suspecting is going to ad a discrete 20 Khz noise line), or Equal Loudness Contour (which will ad a extremely low level of white noise across the spectrum).

Could someone please tell me at least if I'm on the right track.

I really want to restore my vinyl in 24 96, as it will give me much more head room for the restoration processes.

 

Thank you for resurrecting a technical thread!

Just a couple of things for now: You do not have to use noise shaping. There are benefits to using it, but there are benefits to not using it. The current wisdom these days is to not use noise shaping with your dither because it makes the sound worse if you ever have to process an already-dithered file in the future. I personally do not think this is a big problem, but your mileage may vary. However, noise shaping can yield a quieter sound because you can "shape" where the noise will go.

If you want headroom for processing, record at 32-bit float, or convert your 24-bit files to 32-bit float before you process.

My method for choosing the best dither is to do A/B testing with the source file and the dithered one to see which sounds closest to the source. Of course, you will have to take time to experiment. I am always re-evaluating my settings. Oh, make sure you do the sample rate conversion to 44.1 before you dither.

Does Sound Forge 9 use MBIT+ sample rate and dither? I have SF 10, so I don't recall. You may get better results using the MBIT+ options, but, maybe not.

Use speakers, not headphones when doing your experimentation. Headphones will fool you.

 

Indeed. In fact, I recall reading some comments on one of the mastering forums a couple of years back from former Motown engineer Bob Ohlsson, who felt no shaping sounded the most transparent to him.

Another factor is end use. If you're going to be converting files to lossy compression where the shaped noise may be removed altogether, you're better off using no noise shaping (Of course one could argue that the audible changes made by lossy compression are greater than removal of dither noise but that's for another discussion). I thought this might be the case for awhile and iZotope's dither guru Alexey Lukin confirmed it in one of the threads around here. So if end format is to be redbook CD or higher quality then noise shaping is fine, but flat triangular dither is better for any material that will also be converted to lossy be it MP3, or AAC.

I also use flat dither when converting to 24-bit word length after processing needledrops as 32 bits. It's below the audible limits but helps eliminate any distortion artifacts that may creep up into the audible range. Plus as you say, if you later want to do any processing, flat dither is much easier to deal with.

 

I'm pretty sure Bob Katz had similar sentiments. And I myself have never really been a fan of the sonic results of noise shaping, ever since I first heard it on Sony equipment versus no shaping at all. In my experience it can really hurt classical music, especially the tone of massed violins where the sound can become hardened, steely and the effortless pace and timing of the 24 bit master is lost - and the more aggressive the noise shaping the worse these effects are.

I used MBIT+ for a few years but about 18 months ago I came across PSP-X Dither which I purchased based on a testimonial by Bob Katz. I've found that the combinations available with this plug-in can produce a sound much closer to the 24 bit master than any result I ever achieved with MBIT+, no matter what combination I tried. The differences are still audible, but I have more flexibility to the point where my resultant CD master files are a sin of omission versus the 24 bit rather than containing any obvious artefacts that I would consider sonically objectionable. Mind you, I still use Izotope for the resampling as with a lot of patience and tweaking of the various settings, the results can be and are usually excellent. I just never achieved this with MBIT+ unfortunately.

 

Thanks for reminding me! Many pros are using PSP-X in the studio these days, too. When I get some disposable income, i'll have to check it out. Good thing I have been saving mt 24-bit files.

 

I am absolutely against dithering in any form

 

Are you sure you know what dither is and how necessary it is?

 

Grant, The way people miss use the language when talking about digitization here, it would not surprise me if many people are confused.

Jason, I do not mean to pick on you, but, you do not dither down from 24 bits to 16 bits, but rather you say "when reducing quantization levels from 24 bits to 16 bits" or simply say "when reducing bits from 24 to 16". If you are changing both the number of quantization bits and the sampling rate then the process is called decimation.

Dithering always refers to the process of reducing the impact of quantization error on low amplitude signals.

 

Yes. Dennis is likely thinking that dither is like noise reduction.

Here is a nice little two-part tutorial on dither and quantization noise. This is part one:

 

Part two:

 

Here is the example of dithered (clean) and non-dithered (distorted) 16-bit PCM audio stemmed from a 24-bit source for ya...