Does digital audio work like digital images...ie more bits for highend?

Discredited by who? Show me some published work that has discredited it. Point me to a URL or an AES paper and I'll gladly read it.

I bet I've done more high-res location recording than 90% of anybody here, using a Deva 5.8 recorder and Schoeps MK641 microphones. The benefits of anything over 48K/24-bit are almost nil in the real world I deal with. And I'll put the preamps in the Deva up against anything out there.

I can see going with 96K/24-bit for very esoteric orchestral work, and possibly higher than that for complex sound effects work that are going to be subjected to a lot of intense, after-the-fact processing (particularly pitch shifting), but beyond that... no, I think Brad Meyer's research is valid, based on what I know and have experienced myself.

Note that Brad has no agenda, and is not marketing high res or standard CDs as part of his business. I'm not going to point fingers, but I think the reality is that most people who do promote 192K/24-bit are people who stand to make money selling software for these formats.

BTW, those who want to read an interesting discussion by very high-end pro sound engineers on 96K recording are invited to read this:

http://www.jwsound.net/SMF/index.php?topic=2594.0

This is a debate on the pros and cons of 96K recording for film & TV sound. The upshot is, in the real world, there are no advantages and dozens of disadvantages, chief among them being that very few post houses and mix facilities that can handle anything beyond 48K. I'm still unconvinced there are any microphones and preamps that can capture any useful information over 24K anyway, and I don't even buy that transients, harmonics, or aliasing are factors in frequencies that high. I think there are a dozen other factors that are much more important, like distortion and other problems that are far more disturbing and audible. (BTW, the mikes and preamps do exist -- they're just extraordinarily rare, like the Sennheiser MKH800, which only goes to 50K.)

24 bit recording is a given, and I have no problem with that.

 

My experience when not dithering an already 16-bit file is that I get a gradual buildup of distortion. Wouldn't be better to use dither on a 16-bit file at the risk of cumulative dither?

I recently used to advocate not using dither on 16-bit files too, but that was only if I knew if the file had already once been dithered. OTOH, according to Bob Katz, there is no such thing as previous dither making up for not using dither.

 

Heh! It's late. I sure wish more of us were open to opposing ideas.

 

I think it depends on what you mean by staying in 16 bits. If you make a bit-for-bit copy or you're only editing cd markers there's no need to dither, 'cause nothing changes to the audio data itself (at least, it shouldn't).
If you're doing anything to the file that involves calculations to it, which most software does in 24 or 32 bits, you need to dither back to 16, even though it might seem from a practical point of view that you're staying in 16 bits. (You're effectively going 16->32->16, depending on the software)

Forgive me if I'm stating the obvious here, but I think that's what Barry meant by staying 16 bit.

Hey, more bickering!

(though I do agree that 48kHz should have been red book standard from the get-go.)

 

Please do no make such suppositions. I do not agree at all that intersample peaks can be "avoided by using modern converters and modern software". These have nothing whatsoever to do with the vintage of the converter or the software.

Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Hi Grant,

I still don't see the point in dithering a file unless you are shortening the word length.

Are you talking about performing some sort of process on a 16-bit file? If so, this is someplace I'd never go. When given a 16-bit source, I save a copy at 24-bits, then process the 24-bit file. If the target is 16-bits, say for CD, I'll apply dither but that is for a 24-bit file being converted to 16-bits. Not to a 16-bit file.

If this is what you're talking about, then we're on the same page and just describing it differently. What I've described is not dithering a 16-bit file though. It is dithering a 24-bit file (which may or may not have been created from a 16-bit source). I never apply dither to a file that is not about to have its word length shortened.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Sure there is. That's what the bandwidth limit tells us - that we can accurately re-create any signal who's bandwidth is in that limit.

My point is that it's how we limit that bandwidth that is the issue, and going to a higher sample rate allows you to limit that bandwidth with less audible effects from that limiting.

 

There's no perfect filter though - they all either a) allow through aliases, b) ripple or ring. If you sample at a higher frequency, you can use slower filters that ripple or ring less, or even not at all if you can get the sample rate high enough.

We have exactly the same issue in digital video, that there's no one perfect downsample filter. Downsampling is a sampling process, and hence needs an anti-alias filter. In video, the camera has (or should have) the first anti-alias filter, which is an optical low pass filter. They're very simple and very slow filters, but can't, by their design ring as they cannot have negative coeficients. Once the video is digital, we can apply any electronic filter we want when we downsample, but again, the issue is one of trading off aliasing for ringing. The best downsample filter is none at all, so you can get the best looking results from a strong optical low pass filter, a very high resolution (think sample rate) and a very large / high resolution display (to avoid having to downsample).

The argument against that approach is that higher resolution implies smaller pixels, and they are generally noisier. However, pixel designs improve all the time. This is similar to Lavry's point that there could be a sacrifice in going to a higher sample rate in terms of noise. However, if the noise is low enough to capture the dynamic range of the input signal, then a higher sample rate IS better than a lower one.

 

I was referring (as was obvious from the context) to the problem of "loudness" mastering. That was the only issue I expected agreement on.
If you don't think that is a problem on many modern recordings, then I'll respectfully take it back.

The problems with the software and DAWs was that they only displayed the sample peaks and not the peaks of the complete waveform. That was what led to the problem in the first place since the people doing the recording and mastering did not see if the amplitude of the recreated waveform would exceed full scale.
Using modern software/hardware will of course not solve the problem, but they will at least make you aware of it.

To solve it the people doing the recording/mixing/mastering must keep the signal at proper levels.

 

I have been working with 24/96 recordings since 1991 which is when Chesky started getting involved with hirez converters. As the converters have gotten better, the sound quality has improved as well in the areas of 24/176 and 24/192. Up until we got the Sound Devices 722, my team did primarily 24/88.2 work. Then we noticed on our monitoring and playback systems that 24/176 was a bit better - not as big a jump as 24/88.2 was over redbook but still an improvement. We also have DVD-Audio creation capability now and that has helped us test the positive impact of the higher 24/176 rate. I was talking about this very subject with mastering engineer Paul Stubblebine last weekend at RMAF and he certainly hears the benefits of the higher rates as well. He also likes DSD and doesn't hear the upper band distortion which is something I have been saying is inaudible here. Tim De Paravicini also agrees on the value of hirez although he finds analog to have possibly the best playback. After hearing his modded reel deck, I am inclined to agree.

As for the Meyer-Moran paper, they used at times a cheap Pioneer SACD player that did not pass pure DSD for some of their tests. There is a pretty good overview on it at Audio Asylum. There are also AES published papers by Japanese researchers that found benefits from hirez so the community is pretty divided on this.

When I listen to hirez I hear lots more spatial cues in the music and the instrument tonality is just more precise. People may disagree on this but to say there is a technical reason that hirez does not add value is incorrect.

 

Hi Lee,

I think audibility depends on the system and moreso, on the listener.
I have not and will not utilize DSD in my work because, while I feel it offers much that is better than 16/44 (perhaps an understatement), I have never liked the high top.

Rather than describing it as "distortion" or noise, I hear compression and find it fatiguing. Much the same way as many audio components which don't do dynamics as well in some parts of the spectrum as they do in others, this lends a "character" I do not find pleasant and do not deem anything like "natural".

Lots of folks don't hear this and lots do. So, I wouldn't call what is occurring, whether you label it "distortion" or something else, inaudible.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

In a recent discussion of this same topic with a friend, he asked me to elaborate. He wrote:
"Can you define "dynamic range"? Are you talking about the frequency range? When you say that the additional points are added to the bottom of the dynamic range, which particular parts of the music are you talking about? And, if you record same exact music in 16 bit and 24 bit what exactly are we missing from the 16 bit version?"

Below is my response, which I hope might make the subject a little more clear for those who are interested.

***
"Dynamic range" refers the the range of loudness. It defines the theoretical distance between the loudest sound that can be recorded (in digital, represented as 0 dBFS --full scale) and the quietest sound that can be recorded.

Dynamic range has nothing to do with frequency range. Where frequency defines musical pitch, dynamics define musical volume level. (Though some components do better dynamically in some parts of the frequency range than they do in others. This in part, lends the component its sonic "character". For example, if a loudspeaker compresses dynamics --i.e. cannot reproduce the full dynamic range of the input signal-- in the bass, it will tend to sound on the bright side, since the treble component of the signal is reproduced with greater dynamic range.)

CD's 16-bits has a dynamic range specification of ~96 dB. It can capture *in theory*, all of the sounds from the loudest to those that are 96 dB lower in level than the loudest. This leads some Internet audio "cowboys" to make posts suggesting (sometimes insisting) that music with a 96 dB or smaller dynamic range can be fully captured by a 16-bit recording. (Would that they would inconvenience themselves with some actual first hand experience making recordings and listening to them on a good system.)

Some of 16-bit's issues occur at the bottom of that range, in the quietest sounds, where many instrumental harmonics and many spatial cues exist.

At the lower order bits, resolution is less. The loudest sounds are represented by all 16-bits. Since each bit represents just a little (a few thousandths of a dB) more than 6 dB, if a sound has a maximum level that is say, 7 dB below 0, it will only be represented by 15-bits.

This has led some engineers to believe they are not getting "full resolution" unless they keep the signal in the top bit, hence the preponderance of CDs (especially pop) with a dynamic range of less than 6 dB. Of course, getting "full" 16-bit resolution on a program that has had the dynamic Life squeezed out of it (in order to fit it into the top 6 dB) is of dubious value to those of us who appreciate the real sound of music.

By recording at 24-bits, that same sound I referred to above, where the maximum level is 7 dB below 0, will now be represented by 23-bits instead of 15.

Continuing with another example, this time a pianissimo where the maximum level might be say, 30 dB below the full orchestral forte (i.e. 0 dB), with 16-bit A-D conversion, that sound is represented by 11 bits. (30 = 5 x ~6dB/bit, meaning we're not "using" the top 5 bits for that particular sound. 16 - 5 = 11) By using 24-bit conversion, that same sound is represented by 19-bits, still having more resolution than the 16-bit version at full level.

11-bits will sound quite coarse, harmonically thin with little sense of the space the players are in (assuming it is captured by the microphones). 24-bit recording eliminates these problems. This is why astute listeners will hear better reproduction of instrumental harmonics and better "focus" on the space the players are in when listening to the 24-bit recording over the 16-bit recording from the same microphone feed.
***

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

I'm talking about processing. Most software nowadays has at least 24-bit processing. So, it seems to me that one would want to use dither.

 

Exactly, yes.

 

Hi Grant,

I believe even the cheesiest modern software processes internally at 32-bits.
However, a lot of software will save its temporary files at the word length of the source file. That means even if the software processes internally at 32-bits, if the source is a 16-bit file, there will be a sonic price to pay. With such software (I'd say any app where you are not absolutely positive this is NOT the case) you need to do a Save As first, to a longer wordlength, like 32-float, if you are going to REALLY process at 32-bits.

Know too, that most software isn't bit clean even at 24-bits. It is just a digital version of the numbers game some folks love to play because it appears to relieve them of having to listen. "See the spec sheet/ this white paper/ such-and-such an "expert" said..."

(By the way, my editor runs at 48-bits, one of my EQ plug-ins at 60 and my "mixer" at 80.)

It seems we're talking about the same thing but using different words. (?)
As I mentioned, I'd never process a 16-bit file, regardless of what the internal engine of the software processes at.

With a 16-bit source, I'd first do a Save As to at least 24-bits (depending on the software), then process the 24-bit (or greater) file. ONLY if I needed to go back to 16-bits, say, for a CD master, would I apply dither.

With a greater than 16-bit source, say, 24-bits, the only time I would apply dither (and only once) would be as the very last step, after all other processing, level changes and sample rate conversion is completely finished. And again, this ONLY if I needed to go below 24-bits for something like creating a 16-bit CD master.

Even the very best dither is not the same as no dither at all. It must be used (in my view) only when it provides a benefit.

So going back to my original statement, yes, I'm sure. If you're staying at 24-bits or staying at 16-bits (not to be confused with ending up at 16-bits), there is no need for dither, so being at 24 doesn't make a difference. Dither only applies when word length is being changed to a smaller value.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

I hear a lot of things others don't but I guess don't hear the DSD distortion. To me, DSD has an analog like quality and it allows me to listen much longer than 16/44 CD.

On another note, it seems the one thing we discover in these threads is that theory does not match up well with practice in audio. I like theory when it correlates well with practiced recording and playback but audio is very complex and I think theories like Nyquist have their limitations.

 

Hi Lee,

I think Yogi Berra put it succinctly when he said:
"In theory there is no difference between theory and practice.
In practice there is. "

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Quite true.

 

Barry, would you dither when going from 32 bits to 24 (if this were the target bit depth) with 'transparent' software such as iZotope's?

 

Hi Metoo,

In the past, when I was using a software package that saved its intermediate files at the word length of the source file, I would do a Save As on everything when I started a job, in order to have 32-float files to work on.

There was a time when I was dithering when going from 32-float to 24-bits. (At the time, I was using Pow-R dither.) Then one day, I did a comparison with the dithered file and one where I simply truncated to get to 24. The key for me in making such a test, is which "converted" file sounds the most like the long word length original? From that day one, I never again used dither to get to 24.

Keith Howard wrote an interesting article on the subject for Stereophile. While I got the impression he doesn't like dither at all, reading the article stimulated the experiment with truncating for 32 to 24 conversions.

I use MBIT+ whenever I reduce a file to 16-bits but as I've said elsewhere, there is no dither like no dither. For a 24-bit target, I don't use dither anymore.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Yeah, most software processes internally at 32-bit or 64-bit. I mentioned 24-bit because it is the very least, and there are plug-ins that do, or claim to process at 24-bit. I was trying to be as accurate as possible.

For a while, I was processing already-dithered-to-16-bit files with no dither, because I was thinking that the previous dither would take care of it, until Stefan convinced me otherwise, and I re-read Bob Katz book where he states that there is no such thing as files "dithering themselves". AQnd, you even mentioned this. So, I started opening up all my files in 32-bit again to work on them.

Another reason I though it may have been OK is because I was using 1 bit of dither. I have realized that this is too much. See, in the old days, Cool Edit defaulted at .5 bits of dither. When they turned into Adobe, they upped it to 1-bit, for some reason, claiming it is optimal. Well, through some lengthy experimentation, I found that their original default was more transparent without the noise shaping. Now I feel like I am getting much more transparent 16-bit files again. And, for my software SRC, I am firmly in the no noise-shaping camp.

But, i've just digressed again...

 

Again, for a long time, I did not dither going from 32-bit to 24-bit if I ever had to. I didn't see the need either. Then, I read somewhere that it was recommended even though the noise floor was too low for it to matter. I have been doing it lately, but i'm still on the fence about how important it is.

 

Hi Grant,

You can take any perspective on any subject, particularly audio and find an article somewhere that says it is absolutely the way to go.

The way I approached it was to take my own recordings and compare the dithered version and the truncated version with the unprocessed original. My choice for "better" was the one that sounded most like the original. Once I did that, I stopped using any dither when going from 32 to 24.

And this, even though the MBIT+ dither I'm now using sounds to my ears, a whole lot better than any of the other dozen or so dither/noise shaping algorithms I've had on hand.

The software I'm using now doesn't require me to do a Save As to 32 in order to get the best out of it (i.e. it does not save its temporary files as the word length of the original file). So, my 24-bit recordings remain at 24 throughout and copies of these are dithered only for the CD release.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

I was talking to Kevin Grey last week. I did not bring up DSD, however we were talking about sample/bit rates beyond Redbook. He berated the top end quality of DSD and said he thought even the lowly Redbook had a better representation of top end when compared to DSD.

 

That's just a silly opinion imho. If you listen to a mic feed and compare it to a DSD version and a 16/44 version, the deficiencies of 16/44 are pretty obvious.