should I dither? (I can't make my mind up (groan...))

Hi all,
I'm recording some vinyl on pc for cd transfer. I'm using a Teck 1200 and going into the PC via USB. I'm recording this one album at 96/32(float)  but will drop the bit rate down to 24 as to play on my dvd player, ie, audio dvd. My question is with the sample rate so high do I really need to still dither? I was planning on using L2 since its pretty simple and I'm no dither expert. Thanks for any help in advance. PS. I know I can burn it to cd and none could tell the difference but I like the idea of a higher sample rate and bit rate for certain albums since I highly doubt these albums will ever be remastered. I wish to preserve as much of the original signal as I can.

to dither or not relates to bit rate not sample frequency

digitizing is an integer process, means no appreciable difference in the A/D process between 24 bit integer and 32 bit float.  As no commercially available Audio Cards support 24 bit S/N in any case the need dithering while migrating from 32(float) to 24 (integer) probably does little to improve audio results, though theoretically would be hard to detect either way

and the choice is thoroughly, completely, aesthetically, magically subjective

[magically subjective]

and the choice is thoroughly, completely, aesthetically, magically subjective

I like that!

It need not be subjective. The processing can be done both ways, then one of the freeware ABX programs can be employed (e.g. WinABX, PCABX, the foorbar2000 ABX plugin) to show yourself that you cannot tell which is which. You can then go on to downsample to 16/44.1 and test that against the 24/96 version to show yourself that it isn’t possible to hear which is which.

You could also compare recording at 16/44.1 to recording at 24/96 to reveal that you cannot tell any difference there -- except that no two plays of an LP are ever exactly the same, so you may be able to detect which is which based on this factor (unless you can make both recordings at the same time). The same factor would be at work comparing two separate recordings at 24/96.

It need not be subjective.

You could also compare recording at 16/44.1 to recording at 24/96 to reveal that you cannot tell any difference there -- except that no two plays of an LP are ever exactly the same, so you may be able to detect which is which based on this factor

I don't really wish to quibble the point of a throw away line, but even by your own assessment, or statement anyway, the efficacy of dither is subjective.   Basically, even by your own admission the results of dither are masked by the 'noise' of the process.

In blind A/B tests whose results I've looked at individual reaction to dither is nearly statistically neutral.  I have not seen any involving migrating from 32float to 24 integer (when the recording is done @ 24bit) but combined with info I have read and what is going on in the dithering process the only reasonable conclusion is that detecting a difference between dithered and undithered (32float-24integer, when recorded @ 24bit) is subjective.  The suspicion that in such a test that results would be scattered among possible responses (results matched by other dithering tests) basically confirms the subjectivity of the response.

When the results are subjective one can come up with any result imaginable. A long history of research with perception has shown that belief and expectation can be more important than the signal in determining what the subject believes he hears (or sees or feels, etc.). To say the results are subjective, especially in this context, is to say they are imaginary.

ABX testing prevents expectation or belief from influencing the results. If the score says neutral, that is a very strong indication that the person can not hear any difference, which is much the same as saying there is no audible difference, at least within the hearing acuity of that person. I don’t understand your point.

Of you are referring to the “no two plays of an LP are ever exactly the same” part, I say that is totally non-subjective. It is verifiable by comparing the recorded data. If the sound  difference of that data is enough to distinguish in ABX tests, then there is an actual audible difference. This is not subjective unless you want to position yourself in a metaphysical universe that admits to nothing being objective.

One might claim that those who object to ABX testing, once they find that their cherished beliefs are not supported by their test results (i.e. “these cables make a big difference” tests out as “I can’t tell these cables from any other cables”), are influenced into the subjective result of false negatives in ABX tests by their objections. That might be a difficult hypothesis to test.

When the results are subjective one can come up with any result imaginable.

No it's not.  While something imagined without tangible cause/effect relationship might be 'a' result of a subject decision, not all subjective decisions are willfully imaginary by any means.

The point about dither when migrating from 32(float) to 24(integer) (when an audio file is digitized via a 24 bit integer process) was that dithering does something concrete, verifiable to the 32 bit file.  The file is no longer congruent with the un-dithered file.  This aspect is no different from migrating from 32(float) to 16(integer).  The 'something' done is not imaginary.  Unfortunately for at least the special case of 32-24 ability of detecting the audio impact tends to be subsumed, tectonically, by the noise of the system (hardware, software, human detection, etc. . . . entire 'system').  Even if an individual can detect the specific effect of dither/noise shaping in on file that has been processed from 32(float) to 24(integer) the dither artifact is inherently compromised by the noise of the system.

Evidence for this is a statistical, not individual event.  Indicating that, in my experience, with regard to dither, generally, results of WinAB tests are 'statistically' neutral . . . says nothing about individual response what so ever.  I have not sponsored any dither tests but have participated in a fair number of mic 'shootouts' which not surprisingly gives me some idea of limitations of objectivity associated with such endeavors.  Additionally having done collegiate/university training primarily with organisms (& parts of organisms) far too small to detect directly I actually am aware of how difficult it is to construct statistically significant double blind a/b tests. 

And rather obviously, ignoring entirely issues of playback system (used in tests) & influence of training on any subjects, if your statement concerning no two discrete plays of an LP ever being the same is true, until the variables of that 'system' are identified & quantified pretty much no test involved in that audio can more then nod in the general direction of objectivity.  The noise of the system is simply to great.

While in a very general way most of the PC-ABX tests I've perused tend to support, in a general way, most of my attitudes and opinions about what processes do and don't impact audio (and hopefully you can pick up on the ironic tone to the above and one reason why such tests, even if they support things I believe, remain objectively questionable) a great many, that i've perused, lack sufficient double blind rigor to be statistically significant.  Which does not imply that any (and all) claims (including ones that unicorns, while singing in tones that can't be detected auditorially, do cause one to live forever) are supported by the tests, merely that in the dialectic arc between objective/subjective they still tend to be tilted (in a pin ball sort of way) in the direction of the subjective!

One last pointer in all of this is purely anecdotal, & personal.  One of the primary reasons why I don't mind listening to music via MP3s even encoded at relatively low bps (down 2 128 perhaps, though for clients with play back systems that I know will support it/them tend to use command line LAME variable bit encoding or FLAC, both of which tend to only generate 4-1 data compression) is that after more then thirty years it is virtually impossible for me to listen to music with the naive enthusiasm I had at 16.  Analyzing the composition (process as well as performance), performance, mix, speculating on tools and processes employed (were the choruses flown in, was the solo/lead comp'd, is that a U47 @ 18 in. and 44a, with null facing source, @ 3 ft.?, etc.).  While I have significantly less experience working with film my significant others tend to hate attending movies (or live theater) with me because it is difficult to impossible for me not to deconstruct the 'process' with regard to camera angle, lighting, how the length of a particular shot propels the narrative, etc.  I still enjoy music (live music far more then recorded but even there I tend to either want to play with the performers or instead of them . . . & that colors the listening experience) but even when attempting to forcefully suspend disbelief I can't, not hear music the way I now do.  I do not posses 'golden' ears, am constantly surprised by the level of detail that people I work with can detect that I still miss until they point things out.  I'm old enough, of course, that there are some things (and of course always were) that I simply can not hear.  The breadth and depth (however broad or shallow) of experience 'colors' everything I hear.  I do not see this as a bad thing.  But it does make my objectivity, even in double blind A/B, suspect.  And aesthetically 'what' I hear, even in a doubleblind a/b, might or might not have anything to do with dither . . . double blind, by and large should be a statistical process and be applied sparingly to individuals

now I'm bored

[not]

We are just talking at cross purposes here. You might notice that I wrote that the ABX tests, properly done, will show one that it is not possible to tell any difference. I might be wrong about that in some case, but so far the preponderance of the objective evidence supports that view. The exception being what you’ve already pointed out, dithering variations are sometimes audible, between tracks where there is no music signal -- but only if the volume is turned up to levels that are unuseable when music is present.

There are also other circumstances where the data is different, but discriminating between the two by listening is often not possible -- such as with an mp3 versions vs the original source. With flac, there isn’t any difference from the original.

So, someone deciding that there is an audible difference in the matter under discussion, without any objective evidence, i.e. just from subjective impressions,  is almost certainly imagining it. The difference heard is almost certainly going to be shown to be imaginary if the subjective evaluation is followed up by a properly done blind A/B test. The test choices will not differ significantly from random guessing, which is logically concluded to be random guessing. No, I am not saying that subjective, in the general sense, is imagination.

I understand the point about difficulties with statistical analysis of test results. However, one subject, doing the test correctly, will be able to get very reliable results for himself, regardless of any rational difficulty applying those results to the general population. When the audio is on the computer, and one uses one of the available ABX programs, there are only a few rules to follow to avoid screwing up the tests.

One might then argue that an individual’s particular results could be from some factors only temporality effecting the person (taking test while running a 103 degree fever), but if such a worry is present, the cure is simply retaking the tests under better conditions.

On a lighter note, I just can't help chuckling about the title of this thread, "Should I Dither...?"

Dithering about dithering...

No, I am not saying that subjective, in the general sense, is imagination.

except that is pretty much what you were saying:

When the results are subjective one can come up with any result imaginable.

We are just talking at cross purposes here.

No, I'm not talking at cross purpose.  Though I quite rightly after glancing at the post, exited the thread, exited the sub forum and decided to get on with my day

but even more detrimental then obvious true believers with regard to efficacy of forums like this, subject of Steve's high dudgeon with humorous pusedo-rant of an SOS oped piece (well it wasn't  really the subject of Steve's response but . . . .) are the artificially 'objective'.   Those hearty folk who post shit that certainly sounds reasonable, certainly to naive and inexperienced, but on close inspection lack foundation, are based on misconstrued subjective interpretation of out of context items they have 'seen' on the web.  That this class utterly believes what they post (transactionally, at the moment they post it) and post with conviction is harder to defend against then a poster who has obviously drunk some marketing kool-aid. Though a reasonable test, a subjective one of course, for the purveyors of the pseudo-objective is how rapidly they shape shift, prevaricate; 'i didntsaythatjusttlkngatcrossurposesplittinghairhere' continuing of course with utter conviction of belief, merely modifying presentation of the 'message'.

as the OP's task is far more AndyH's area of experience then mine this (post) and last two are certainly tempest in teapot dangling minutia, but understanding, being able to evaluate objective/subjective, the usefulness of both ends of the arc does speak to the value, if any, of this type of anonymous peer2peer forum

yada yada yada, espresso is getting cold

Well I wasn't going to say anything until I got 'invoked'...

But dithering 24bit signals? And exactly at what level does this put the dithering? (I'll give you a clue; -144dB!) And precisely which sound device can reproduce down to that noise floor, or anywhere near it? On a top-flight converter, the noise floor is around the -115dB mark, and you'd need some serious liquid cooling to get it any lower.

And whether you dither or not has nothing at all to do with the sample rate (the rather excessive 96k in this instance) - it's only a bit depth issue.

Dithering is only worth doing on 16-18 bit signals.

And in that case it's necessary after any processing of any kind; strictly, whenever a signal with more bits is reduced  to 16/18 bits (or less!) - it's meaningless unless you start with more bits than you end with - dither improves the accuracy of the result.  You could think of it as rounding for audio signals (though it isn't).

Paul

... it's meaningless unless you start with more bits than you end with - dither improves the accuracy of the result. 

Whilst it's obviously true that correct dithering improves the accuracy of 16-bit results, and that often this is clearly audible at the end of reverb tails, it's not entirely true that it's meaningless without the greater bit depth to start with, although the exception is pretty rare, admittedly. But, if you end up with signals from one of those horrible old 16-bit converters that didn't dither the least significant bit, you quite often ended up with this bit hopping down the last step between bits 15 and 16 after the end of a reverb tail, sounding a bit like you'd switched something off. Just applying dither in this situation, even though it doesn't correlate with any signals, improves things no end. OTOH, so does adding any old bit of noise at about -85dB! Those converters really did make a mess of a good input signal; I remember feeding Revox PR99 signals via outboard dolby NR to one, and hearing distinctly inferior results. Put me right off for a while!

Dither effectively eliminates quantization distortion. That is the one and only purpose I’ve ever read about. Is there another? Perhaps, based on SteveG’s latest post, we should add that the noise itself, by a somewhat different process, can mask certain problems?

Dither comes in many flavors and amounts. When converting down to 16 bit it is possible to use dither that is itself audible in lower level passages. I don’t know if that is true about converting from 32 bit to 24 bit as CE2K doesn’t give me any control over what it does. However, the real point of it dithering not making any difference in the instant case is that the quantization distortion, if any, will be so small that it could not be audible.

That is, besides the fact that there is so much broadband noise coming off any LP that, even though it isn’t ideal dither, it alone is adequate to assure there is no quantization distortion.

When the results are subjective one can come up with any result imaginable.

To say the results are subjective, especially in this context, is to say they are imaginary

.
oretez,
I guess it is not completely unreasonable to have taken the first quote as applying to listening in general rather than just to the kind of threshold discrimination involved in this thread. My other sentence is indeed poorly constructed and doesn’t read as I intended. There is no one to blame but myself for any misunderstanding. However, I still don’t understand the reason for the controversy that began with my first post.

I wrote, do some ABX test and convince yourself you can’t hear any difference either way, at well as can’t hear any difference with some other listed variations. I don’t see how that get interpreted to “the efficacy of dither is subjective” which I did not say. I didn’t make any reference to other than certain specific circumstances, which is hardly a cause to jump to the general condition.

Was that misunderstanding your whole reason for the attack? In your first post you went on to say that your experience, and the published data, lead you to the conclusion that one won’t be able to hear any difference, no? This seems to me not to differ from what I had just written except that you seem to be arguing against anyone using ABX tests to gather evidence about their own experiences.

It might be somewhat helpful to have a look at history, where dithering initially came from. An important part of DSP are digital filter and feedback control structures. While the pertinent algorithms were developed very early (around 1970) the hardware with sufficient resolution was not available then. Everything having more than 8 bit was extremely complex and expensive. Very soon it turned out that circuits such as IIR filters or control loops had a trend to develop periodic signals of one or a few bits amplitude at the output, though no input was present. This was mainly due to the limited resolution. Such signals were pretty annoying as they were audible in telephone or produced vibrations in a control system as they occupied distinct parts of the spectrum. To avoid that, a small amplitude random signal was introduced that killed the periodicity otherwise present. This was then called dithering.

Those people that used digital pen plotters often possibly remember that these machines had the trend to produce some humming at certain pen positions when left idle. With respect to the current problem of transfering LPs to CD, there is already more than sufficient dithering present due to the inherent surface noise. Dithering is already done !