Easy Question - Volume

[okydoky]

Originally Posted by dcollins

The DAC just tries to keep up.


DC

Originally Posted by dcollins

When you are talking perception and not measurement, it is open to interpretation.

Sometimes you (u) may read something like 'a decibel is the smallest audible increment' which may be true for an average group of people, but many can hear much less that that.

It's also said that for something to sound 'twice as loud' you need to increase SPL by 10 dB, but that is also subject to interpretation depending on what sound and what listener.

Of course intensity is a also measure of work, so it's going to include energy over time (and area).

DC

Originally Posted by Lupo

Hi!


The first big averager is the output reconstruction filter. This "smears" dozens of samples together, if not more. Even if the DAC have lots of ultrasound output, some gear along the path is going to limit the maximum response time to something close to 20KHz. The next big averager is the hearing system. The integration time of the ear means that we hear at least a three digit number of samples at once - about 100 samples of 44.1KHz is about as simultaneously as the ear ever gets. Most often, we're hearing a lot longer events.

It's always a bit of a crisis in the start/stop phases of a sample train intended to be continous. The dither probably suffers in the very first samples, but so does the filtering! It's all bit wack in the start. It soon settles though, it takes a a couple of dozen samples or so to get the sample train going steady enough. By the time the hearing system makes a response, it's settled into thousands of samples.

Hope this was somewhat within the scope of what you had in mind. Please bear with me if you feel it's off topic.


The analogue'ifier effect of dither is the same for the Y axis of sampling as the filter is for the X axis. Dither smooths level, filtering smooths time. It's still a sampler storing digits, by all means, but the squareness is gone. It trades the limited set of absolute values in a truncated system, for (with TPDF, literally) a bit less absolute valued numbers. And.. the nearly magic thing: one or more bits having the uncertainity and endlessness that can be found in analogue. I love analogue. To bring that stuff into digital is, in my view, amazingly neat.

Cool! No need to argue about that, we can call it what we like. It still does the same job for us all. It would definitely be preferable if they could, at last, make it transparent to the end user!


PS: Hope this doesn't come across as hostile in any way.. It sure isn't!


Andreas

hi,

i think it is a mistake to try to generalize about hearing. i am aware of nothing saying that the ear is incapable of perceiving minute variations, or that it simply "averages things".

as david collins noted, different people are going to have different levels of ability to discriminate.

one thing to take into account is the fact that, if we simply stop giving people an oppertunity to discriminate and function quickly, then people may even tend to lose those abilities [i.e "if you don't use it, you lose it"].

with all the compressed formats and so forth there does seem to be a willingness to sacrifice any semblence of quality for portability and convenience. i think that may be a disservice to the public sometimes.

here is a link:

Sensitivity of Human Ear


it seems to me that david collins was saying that the dac tries to "keep up", and that was my understanding. i don't think dacs are routinely designed to deliberately average groups of sample for output, but perhaps there is somethiing i am unaware of in that regard.

i would agree that there are plenty of places in the audio chain where things do not work as well as they theoretically could. but one cannot use deficiencies in one area to argue that it is useless to try and improve other areas.

i am not sure what you [lupo] are talking about when you talk about "response time" being limited by some equipment's bandwidth limitations in the frequency domain. i don't see that as relevant. amplitude and frequency have to be considered separately in this context. we are talking about "averaging of amplitude quantization over a period of time by means of dithering" versus "pure analog resolution in the first impression, without quantization or averaging".

i think dither is useful in alleviating quantization distortion. i do not think it is necessary, or even truthful, to say that it results in the exact same thing as unquantized analog audio. it is a process, and it does what it does. it is generally viewed as an improvement over the alternative. for some reason there seems to be an emotional push to convince people that dither results in perfection.

it can be easier to make use of digital audio once one is o.k. with the fact that it's implementation is not so perfect.

i think it is better to avoid word length reduction altogether, to whatever extent it can be avoided, and i think we would be better off if cds had a higher bit depth. i would rather see a push toward that goal.


right.

[dcollins]

Originally Posted by oky****

i think dither is useful in alleviating quantization distortion. i do not think it is necessary, or even truthful, to say that it results in the exact same thing as unquantized analog audio. it is a process, and it does what it does. it is generally viewed as an improvement over the alternative. for some reason there seems to be an emotional push to convince people that dither results in perfection.

It's not based on emotions, or a multiple choice question. Properly dithered, the amplitude and time-accuracy is "infinite." What's not to like?

DC

[okydoky]

Originally Posted by dcollins

It's not based on emotions, or a multiple choice question. Properly dithered, the amplitude and time-accuracy is "infinite." What's not to like?

DC

hi,

i think you know that it is not "infinite" in an unqualified sense. dither requires a length of time and a number of samples to achieve a statistical effect. there is a reluctance to include that fact in the discussion. there is also ongoing controversy about when, how much, and how to to use it, as well as the obvious limitations and the fact that it is not necessarily going to be perfectly implemented. the nature of the signal to be dithered also has some impact on its efficacy.

i'm not sure what you mean about "liking" it. i have no problem making use of it. i guess i would "like" it better if dither were not needed. it just seems that people [not just you] are not satisfied with it as it really is, so they want to paint is as something better. i don't really see any benefit it that. it doesn't make it work any better. nor does it improve the signal to noise ratio. currently, a higher noise floor is one thing "not to like" [and i obviously understand that quantization distortion generally sounds worse than dither].

i still say we're better off pushing for a higher bit depth delivery medium.

[edit]
and before someone gets all twisted for no reason, none of this should be taken the wrong way by anyone. this is just a discussion, and i'm just stating a viewpoint. i think we should at least be able to get a 24 bit delivery medium going at this point. what's the ***king holdup? [edit]

right.

[Nordenstam]

Hello!

Originally Posted by oky****

i think it is a mistake to try to generalize about hearing.

Possibly. Don't tell those who spend their lifes in the field of psychoacoustics! :D

Originally Posted by oky****

i am aware of nothing saying that the ear is incapable of perceiving minute variations, or that it simply "averages things".

The hearing system is the most amazing instrument around. It's ability to gleam information from small details never cease to impress! I'm the first one to argue that we can hear extremely small details. That is partly due to the special way the ear<>brain works. Unlike audio sampling and regular test instruments, the hearing system is not a linear device giving precise microsecond to microsecond readings of amplitude variations. It's far more complex than that. What I had in mind was a few, perhaps subtle, perhaps obvious, things. First; practically speaking, a sound have to be "measured/sampled" across some stretch of time if it is to have any power. An instantaneous event have no power at all, as it doesn't have any time to deliver that power. The ear needs to spend some time "reading the signal" before anything much at all happens. Logically, there also needs to be a larger piece of time to juxtapose a small event against. Even if it's an impulse, the ear will be using the lack of succeeding event after the initial blast to determine that it was indeed only a single blast. Taking a longer piece of time into consideration is part in determining that it was a short event. Which leads on to the important point. Once the ear gets started, when it receives a transient/onset of sound, it'll integrate anything of the rougly same type of event comming up to 20-50 millisec later as part of the same sound. The Haas effect. As can be witnessed in the use of chorus boxes. In some circumstances, far longer time windows are in operation. Loudness increases with increasing duration up to about a second.

An example of this is the difference between a single and repetetive pulse. The lone impulse will have the full frequency range of the system, its impulse response. If the impulse is repeated, it'll be periodic waveform with a fundamental and harmonics, a low duty cycle square wave. The discrete impulses is lumped together as a single piece of information. The earbrain likes to do that.

Originally Posted by oky****

it seems to me that david collins was saying that the dac tries to "keep up", and that was my understanding. i don't think dacs are routinely designed to deliberately average groups of sample for output, but perhaps there is somethiing i am unaware of in that regard.

To "keep up" - doesn't that mean it have to accept and convert whatever it receives?

Output filtering is doing averaging across samples. This is a single sample:

Upper row shows connect-the-dots sample view. Lower row is the same lone sample, low pass filtered. Notice the pre- and postringing of the sharp lowpass linear phase filtering. The ringing extends a couple of dozens of samples before and after the sample point. The final output waveform is the superposition of a bunch of such filtered impulses that all stretch beyond each sample point.

A bit more illustrations and talk about this can be found here and a longer better explanation is here.

Originally Posted by oky****

i would agree that there are plenty of places in the audio chain where things do not work as well as they theoretically could. but one cannot use deficiencies in one area to argue that it is useless to try and improve other areas.

Though it was not my intention to argue what you describe, it's a good point! Think it relates to the issue at hand. If you separate the whole dither thing from the noise itself, I think you'll find your issues with the way dither works to be found in noise synthesis in general. An impulse contains all frequencies, but only if it's not repeating! To make noise, the signal have to have all frequencies and be non-repetitive. A set of samplepoints representing noise can be seen as all sinewaves in the world mixed together. The only way to have perfect sines and therefore perfect noise, is to have infinite time. At any give stage in the noise output, you'll not find every and all frequencies across a small amount of time. There will be some of these, some of those, changing from an instant to another. As long as the fourier response is flat and it doesn't repeat, it'll be as random as random gets in audio. The same situation applies to reverse zeners and other analogue noise sources.

Originally Posted by oky****

i am not sure what you [lupo] are talking about when you talk about "response time" being limited by some equipment's bandwidth limitations in the frequency domain. i don't see that as relevant. amplitude and frequency have to be considered separately in this context. we are talking about "averaging of amplitude quantization over a period of time by means of dithering" versus "pure analog resolution in the first impression, without quantization or averaging".

What I meant is that any and all signals are subject to the same averaging processing I describe. Both due to lowpass limits in the equipment(not so important) and the way the earbrain works(important!).

So this may very well be a another case of us talking about the same thing from diffferent angles! Aint it typical?

Originally Posted by oky****

it can be easier to make use of digital audio once one is o.k. with the fact that it's implementation is not so perfect.

Agreed. Though I don't think dither is to blame as long as it works.

Originally Posted by oky****

i think it is better to avoid word length reduction altogether, to whatever extent it can be avoided, and i think we would be better off if cds had a higher bit depth. i would rather see a push toward that goal.

Concur, again! =)


Andreas

[Nordenstam]

PS, here's a picture of 24 bit, dithered and truncated waveforms. The upper rows shows about 40 milliseconds, the lower rows shows ~3 milliseconds. The yellow line stays on the same sample point in both views. Notice that this waveform display (in Izotope RX) show intersample peak information.

[dcollins]

Originally Posted by oky****

there is also ongoing controversy about when, how much, and how to to use it, as well as the obvious limitations and the fact that it is not necessarily going to be perfectly implemented.

As I have said before, it is not remotely controversial. The level, type, etc. has been both established mathematically, and proven by measurements and listening. Where do you get the idea that it's somehow in question? Is this in Bob's book, or from the Internets, or something?

Have you read Vanderkooy and Lipshitz AES paper?

i'm not sure what you mean about "liking" it. i have no problem making use of it. i guess i would "like" it better if dither were not needed. it just seems that people [not just you] are not satisfied with it as it really is, so they want to paint is as something better. i don't really see any benefit it that. it doesn't make it work any better. nor does it improve the signal to noise ratio. currently, a higher noise floor is one thing "not to like" [and i obviously understand that quantization distortion generally sounds worse than dither].

It's needed, period.

i still say we're better off pushing for a higher bit depth delivery medium.

Which would still need intermediate results to be d*thered.

DC

[kheftel]

Originally Posted by Waltz Mastering

...who is not 100% deaf.

Yes, I guess to a deaf person, there is no perception, so there is no concept of loudness.

[okydoky]

Originally Posted by dcollins

As I have said before, it is not remotely controversial. The level, type, etc. has been both established mathematically, and proven by measurements and listening. Where do you get the idea that it's somehow in question? Is this in Bob's book, or from the Internets, or something?

Have you read Vanderkooy and Lipshitz AES paper?

hi,

yes, i have read the aes paper, a long time ago. i believe it is currently in storage. if there is something in it that you could quote and cite to, that would be helpful.

you do not appear to have any question as to how to implement it, but you haven't really disclosed your "formula", so i'm not totally sure what it is. i think i know what you mean, and i'm aware of what i understand to be the "consensus" [flat tpdf at 2x lsb for all intermediate steps, then noise shaped dither, high pass tpdf, or flat tpdf at the very end]. what is your understanding?

but there are definitely a whole bunch of alternate theories, and the software usually has a bunch of options for "dither lite", and all kinds of other stuff. there is also quite a bit of support for not using any dither at all for the intermediate 48 bit to 24 bit stuff. lipshitz and vanderkooy seem to be still doing further work and research themselves, no?

so i do not think it is fair to say that implementation is entirely a settled matter.

Originally Posted by dave collins

It's needed, period.

not necessarily. not trying to be contentious, i just don't think you can make such an unequivocal assessment.

and, in fact, some people say they are not bothered by quantization distortion. some people talk about it giving an "edge" to the music. i know, but that's what some people say. whatever.

i also think it is proper to consider the fact that increased sample rate also reduces quantization noise. 4x sampling apparently yields one additional bit [approx. 6dB]. so dither is not the only way to address the issue.

Originally Posted by dave collins

Which would still need intermediate results to be d*thered.

DC

here i think there is a legitimate difference of opinion between groups of reasonable people. and it is possible to work in such a way that there is no word length reduction involved. some people do that. many people seem to believe that the the "intermediate dithering" stuff is unecessary or even counterproductive.

following is a quote from someone over on the prosound site, as an example of some varying viewpoints. its not my post. you may find it interesting. i'll emphasize some portion of it that may be interesting to consider.


"hi dave; <edit> never mind, bruno said it, i think we are
straight on the purpose...

maybe there is a case to be made for permitting some quantization errors
in the interest of keeping noise levels in check on a 100 track
project that is coming out of a daw for further mixing?

most daws have a mix bus that is more than 24 bits; we would
be re-quantizing 100 times; some errors could become
amplified rudely, thus some dithering seems needed.

however, i do hear dither noise building, or the effect of multiple dithers
seems to affect the perceptions of spacial cues that occur at low
levels. try duplicating a float file 100 times, then sum it
with one dither to 24 bits. then dither the 100 files
each to 24 bit, then sum the same way. i think
you will hear a change in the -vibe- which
may or may not be desirable.

imo, the choice is sometimes subjective. depending on the source(s), and taste.

there are "partial" dithers that dither less deeply. (sometimes referred to as "type 2")
imo, these sometimes could be best choice. e.g. if distortion at bit 24 would be
eliminated from the master when the product is dithered for 16 bit formats.

imo, if it seems worth producing a master at 2496, perhaps just for posterity and
future generations, then imo the judgement call is important. so that we will be
hearing details that are supposed to be in the music (that may have been
formerly assumed to be inaudible in lower formats); not "junk" that
was formerly flying under the radar, or "unnecessary noise".

to return to my former example of a 100 track mix: perhaps an over compressed
snare track would not contain even 23 bits of signal data. dither on such
track, converted 32->24 would not be effective, therefore not needed.

also related imo: i observe that in some cases: some noise shaped dithers have caused
more problems for me than they seem to solve. once a client asked me
to "turn off the exciter". i've told the story before. i just
turned off noise shaping and he was happy.

my strong personal opinion is that noise shaping is a big reason why engineers
question whether they should dither 32->24; a lot of dithers default to noise
shaping, it is another variable; which for me, really screws with the sound.
imo, noise shaping obscures our ability to observe effects of dither.

so imo, only by listening and judgement could we answer the question:
"should i dither -this- 32->24 conversion? and how?"

jeff dinces


cerberus audio services"


in any event, no need to be up in arms about any of this, for me anyway.


right.

[okydoky]

Originally Posted by Lupo

Hello!



Possibly. Don't tell those who spend their lifes in the field of psychoacoustics! :D

hi,

i don't mind that study, but i sometimes find the push to figure out exactly how to trick people into thinking they are hearing something that they are not really hearing to be a little overdone.

Originally Posted by lupo

The hearing system is the most amazing instrument around. It's ability to gleam information from small details never cease to impress! I'm the first one to argue that we can hear extremely small details. That is partly due to the special way the ear<>brain works. Unlike audio sampling and regular test instruments, the hearing system is not a linear device giving precise microsecond to microsecond readings of amplitude variations. It's far more complex than that. What I had in mind was a few, perhaps subtle, perhaps obvious, things. First; practically speaking, a sound have to be "measured/sampled" across some stretch of time if it is to have any power. An instantaneous event have no power at all, as it doesn't have any time to deliver that power. The ear needs to spend some time "reading the signal" before anything much at all happens. Logically, there also needs to be a larger piece of time to juxtapose a small event against. Even if it's an impulse, the ear will be using the lack of succeeding event after the initial blast to determine that it was indeed only a single blast. Taking a longer piece of time into consideration is part in determining that it was a short event. Which leads on to the important point. Once the ear gets started, when it receives a transient/onset of sound, it'll integrate anything of the rougly same type of event comming up to 20-50 millisec later as part of the same sound. The Haas effect. As can be witnessed in the use of chorus boxes. In some circumstances, far longer time windows are in operation. Loudness increases with increasing duration up to about a second.

well, i understand that time is involved in musical activities, and in listening to them. i don't know that the ear always integrates events, or that it does so exclusively. when it integrates, it is also registering the individual events, even if not in a manner which is susceptible to immediate observation or study. the haas effect does not necessarily give support to the dithering claim of "averaged infinite resolution". the haas effect is primarily a mechanism evolved or bestowed to aid in localization. if you play a sound from the same source with a 30 millisecond delay, it would be easily detectable as an echo.

to the extent it invokes a statistical averaging over time, the addition of dither is probably discerned as smear, at least to some extent [i'm not saying it does not have good effect]. and listening to each sample expanded out to hundreds of samples [and thus "linearized"], is going to be different than simply listening to the source without the "statistical averaging". under your [?] analysis, in one case the ear is averaging the actual amplitudes, in another case the ear is averaging a series of collections of averages, each spanning a longer time than the actual original signal's amplitudes.

its "two different things". even if the dithering does have a helpful effect over time, it results in a measurably different thing.

Originally Posted by lupo

An example of this is the difference between a single and repetetive pulse. The lone impulse will have the full frequency range of the system, its impulse response. If the impulse is repeated, it'll be periodic waveform with a fundamental and harmonics, a low duty cycle square wave. The discrete impulses is lumped together as a single piece of information. The earbrain likes to do that.

an impulse repeated will not necessarily be periodic, but i believe i understand what you are saying. and if you mean, "The discrete impulses are lumped together as a single piece of information", i disagree. even if the ear is capable of appreciating the "forest" of the several impulses, it is also keeping up with the "trees".

Originally Posted by lupo

To "keep up" - doesn't that mean it have to accept and convert whatever it receives?

i think david means that the dac tries to decode on a sample by sample basis.

Originally Posted by lupo

Output filtering is doing averaging across samples. This is a single sample:

Upper row shows connect-the-dots sample view. Lower row is the same lone sample, low pass filtered. Notice the pre- and postringing of the sharp lowpass linear phase filtering. The ringing extends a couple of dozens of samples before and after the sample point. The final output waveform is the superposition of a bunch of such filtered impulses that all stretch beyond each sample point.

i am not totally sure what to make of the graphs and so forth. i think your example shows that ringing and smear is a problem, and it seems to me that it is only to be exacerbated by dither.

\

Originally Posted by lupo

What I meant is that any and all signals are subject to the same averaging processing I describe. Both due to lowpass limits in the equipment(not so important) and the way the earbrain works(important!).

i cannot say that i agree with you on that because i believe that the ear is not limited to the averaging processes [although it has the ability to work that way], and i do not believe that lowpass limits in the equipment [i assume you are speaking of frequency and filtering] are limiting amplitude quantization or the ear's perception of amplitude on an event by event basis.

Originally Posted by lupo

Agreed. Though I don't think dither is to blame as long as it works.

i can only say that i do not think it works as perfectly as some may wish. i.e. it does not yield the exact same ouptput as the analog input. it also does not yield "infinite resolution" in the amplitude domain, because that would require an infinite amount of samples. it also does not achieve infinite resolution in the time domain, other than in the limited context of a "statistical averaging" [and also because it would again require an infinite amount of sample, which would possibly tear a hole in the universe because they would all have to occur simultaneously or something. it also seems to me that, even at best, redithering can only achieve the accuracy of what is being redithered.

that said, i have no big issue with it, i just think it is better to stay objective.

there is always interesting information in your posts, by the way.


right.

Originally Posted by lupo

Concur, again! =)


Andreas

[Nordenstam]

Hi!

Originally Posted by oky****

i don't mind that study, but i sometimes find the push to figure out exactly how to trick people into thinking they are hearing something that they are not really hearing to be a little overdone.

It's closely linked to how we sense the world in general. The factual information being sensed is extremely coarse compared to the ideas about the world the brain presents. The post processing of the senses is what gives us the hifi vision and hearing we enjoy.

Originally Posted by oky****

well, i understand that time is involved in musical activities, and in listening to them. i don't know that the ear always integrates events, or that it does so exclusively. when it integrates, it is also registering the individual events, even if not in a manner which is susceptible to immediate observation or study. the haas effect does not necessarily give support to the dithering claim of "averaged infinite resolution". the haas effect is primarily a mechanism evolved or bestowed to aid in localization. if you play a sound from the same source with a 30 millisecond delay, it would be easily detectable as an echo.

The point is that the brain is very apt at filling in gaps. What we sense as noise is nothing but a bunch of different lumps of energy that we later connect as a continous event. We're very good at doing the same with other signals.

Here's a cool page with a bunch of audio illusions illustrated with wave file examples: Demonstrations of Auditory Illusions

The "gap illusion" thing, amongst others, is highly revelant to the brains ability to connect discrete events as a single one.

Originally Posted by oky****

to the extent it invokes a statistical averaging over time, the addition of dither is probably discerned as smear, at least to some extent [i'm not saying it does not have good effect].

The smearing idea seems right. A transient signal mixed with noise will have similar frequencies as the signal of interest at the leading and trailing edges of the transient, due to the all encompassing nature of noise. It's a property of mixing a noise floor with signals - not an effect of dither itself.

Originally Posted by oky****

even if the dithering does have a helpful effect over time, it results in a measurably different thing.

Can you show that in any way? Make some A/B samples or similar?

Originally Posted by oky****

an impulse repeated will not necessarily be periodic, but i believe i understand what you are saying. and if you mean, "The discrete impulses are lumped together as a single piece of information", i disagree. even if the ear is capable of appreciating the "forest" of the several impulses, it is also keeping up with the "trees".

Fire up your favourite synthesizer and twist the pulse width knob on the oscillator. Tell me what you hear! The forest, or the trees?

Originally Posted by oky****

i think david means that the dac tries to decode on a sample by sample basis.

It can't. That's not how sampling works.

Originally Posted by oky****

i am not totally sure what to make of the graphs and so forth. i think your example shows that ringing and smear is a problem, and it seems to me that it is only to be exacerbated by dither.

It's the fundamental sampling theorem. The waveform being stored and reproduced is the product of many samples. It's not a sample-by-sample system. Check the Lavry papers for a great explanation.

Originally Posted by oky****

i can only say that i do not think it works as perfectly as some may wish. i.e. it does not yield the exact same ouptput as the analog input. it also does not yield "infinite resolution" in the amplitude domain, because that would require an infinite amount of samples. it also does not achieve infinite resolution in the time domain, other than in the limited context of a "statistical averaging" [and also because it would again require an infinite amount of sample, which would possibly tear a hole in the universe because they would all have to occur simultaneously or something. it also seems to me that, even at best, redithering can only achieve the accuracy of what is being redithered.

You have to remember that the "infinite" we're interested here is maximum 100dB or so (below the LSB). That's plenty infinite for audio!


Now the above was just to follow up. What made me post this was that I have indeed been taking your posts about dithering seriously. Have spent some time trying to fault dithering. My presumption is that if there's anything wrong with it, it should be possible to create some artifacts in the process. If that can be done - the artifacts can be found too.

What I did was to use extremely low frequencies as test tones and a double highpassed TPDF dither. This dither gives about 100dB noise free range below the LSB, in the bass end. I figured if there was anything wrong to find, using that dither and such frequencies would be a good starting point. After banging my head into the usual FFT limits, the end result was: nothing. Using the most extreme analysis tool I have, the Izotope RX with silly amounts of frequency and time overlaps on the windowing, shows nothing at all. Except for the noise (and potentially FFT artifacts too - watch out!), there's nothing to be found.

I figure that if there is anything wrong, it should turn out as a measureable artifact. It didn't. Please have a go yourself! Would be very interesting if you can find anything but signal+noise in a dithered sample train.


All the best,

Andreas Nordenstam

[lerone]

Originally Posted by jdg

grab a limiter, put it on the master bus.
drop the threshold on it till its LOUD as you want.
set the ceiling to -.2dB or something. export as 24bit.
make the mp3 from the 24bit wav (LAME for example works with 24bit input sources)

for PT, everyone loves this limiter:
High-end plug-ins for Pro Tools

good luck

Yep! The massey is a great limiter for PT. May I add, if you can use a LP EQ to hi pass at about 40hz (come on, its myspace and low pass around 12-13khz, it will sound better when encoded with the evil myspace compression + you will more than likely get a couple of extra dbs out of your track

[kheftel]

Originally Posted by lerone

Yep! The massey is a great limiter for PT. May I add, if you can use a LP EQ to hi pass at about 40hz (come on, its myspace and low pass around 12-13khz, it will sound better when encoded with the evil myspace compression + you will more than likely get a couple of extra dbs out of your track

Wow, that's a lot of frequencies to take out, almost two whole octaves. Darn that myspace evil compression.

[dcollins]

Originally Posted by karyoky****

hi,

i just saw your post, and didn't want you to think you were being "ignored".

i don't know what to tell you other than that trying to manipulate the the frequency content, or other aspects, of music, in the hopes of making up for shortcomings, by trying to "take advantage" of what you may speculate or believe to be the way people hear or perceive is nothing but deception and a degradation of the music.

you seem to think that people [or some people] only hear continuous sound in somewhat larger "chunks" that they put together later on in their heads or something. so you figure that just playing the music in such "chunks" of sound in the first place is just as good as playing music as continuous sound. i believe that is erroneous on its face.

just because the human being may [or may not] tend to parse information in ways that you speculate about does not mean that your presentation of the information to the human being in such a way in the first instance is the same thing as, or that it will result in the same effect or interpretation as would, the presentation of the complete information in the first instance.

what you are proposing is basically the intentional placing of a layer of confusion in the chain.

basically, the less you interfere with what occurs naturally, the better. in my opinion, you shouldn't try to second guess the ear and brain.

i'm starting to think that you may spend a lot of time dealing with artificial sounds ["ambient d j stuff" ?], and if you were more oriented toward actual recording of natural sounds you may have a different outlook.




with all due respect, i don't think you are in a position to say "how sampling works". there are lots of different methods and implementation used by different manufacturers and designers.

to recap, you are talking about david collins' statement that dacs deal with samples on an individual basis. just understant that at some point each sample has to be dealt with, even in devices that do averaging.

judging from some of the papers and data sheets i have seen, some dacs do averaging and some don't.

your idea that people or machines are incapable of parsing information at a rapid rate is erroneous in my opinion. even if someone is not consciously aware of all the detail that individual is parsing, it is still being rapidly done. if something comes at your eye very rapidly, you will blink to avoid injury, even if you cannot consciously track what it is.




i understand the "sampling theorum". it is a sample-by-sample system as well. you always seem to want to trivialize the importance of the individual samples.

i do not find lavry's papers to be all that great of a source of information. those are obviously white papers and marketing tools, not texts. i know you are very much into mr. lavry, and you seem to have a need to make that point repeatedly, but your efforts in that regard are lost on me. you're barking up the wrong tree.




i'm not sure what point you are trying to make, and i do not really understand what you are saying, or trying to say. i seems like you are still trying to continue that long drawn out debate you had with monomer about dither, or something or other.

if you are trying to say that a dithered 16 bit file is the same thing, or just as good, as its 24 bit origin, you are wrong in my opinion.

if you are trying to say that the statistical averaging aspect effect associated with dithering is just as "perfect" as the original sound, i disagree.

there are only a few people that i am aware of who seem to have some huge desire to have others believe that dithering results in an identical thing as the original sound.

even the guys that apparently did the seminal work on dithering in audio do not make that assertion. they just say that dithering can give a result that is statistically the same on average over a period of time. that's cool, but that's all there is to it. and the texts and manufacturers' statement that i have read claim nothing more either.

face it, there appear to be some things about digital audio, and its implementation that are questionable. anyone who does not recognize the fact that certain types of sound, images, or information in general can have an "agitating" effect on people, and most particularly, those who are not so aware, is naive.

the thing that dither does that is actually very useful is remove quantization distortion.

some people seem to be going to great lengths to try to convince themselves of things that make them feel good about using digital audio or something. it is what it is.

as far as any request for a b comparisons or whatever, it is denied. i am not a test facility, and neither are you, to my knowledge. i am always a bit amused by guys who are posting pictures and graphs of this thing and that thing that they have come up with on their personal computers, as if any of that is supposed to prove something.

i'm not sure what you are trying to find out about, but i think audio precision and prism makes test equipment for those interesting in scientific testing. i think bob katz said something about needing a bit scope or something like that to examine digital audio. whatever.

Sorry to quote this whole thing back, but outside of the part where I saw my name I can't tell what is trying to be communicated here.

Are you for testing, or against testing? For the sampling theorm or against it?

We never hear an individual sample, the integration is part of everything from the "sigma" part of the delta/sigma converter to the actual physiology of your hearing. It's everywhere and inescapable.

That's a good thing as it allows you to hear below the noise floor by letting the noise average "out."

I once heard it said the brain is a massively parallel computer, with a clock frequency of 1kHz.


DC

[dcollins]

Originally Posted by karyoky****

i guess i am not really concerned with it in the manner it is usually spoken about here.

See, we do agree!

i almost never see anything to indicate that these people are accredited as testing facilities in any way shape or form, or that they are using any accredited testing faciity, or that they are even qualified to think about it any more deeply than a random guy like me.

The thing is these principles have all been verified using both book-learnin' and "accredited testing facilities."

the sampling theorum is not something to be for or against, is it? it seems to make sense, as far as it goes.

i don't know that it proves quite as much as some would like, but i guess if you had some signal that was 100% periodic, and you took some perfect, evenly spaced samples of it at slightly more than twice its frequency you would be able to figure out what its frequency was. o.k.? that's about all it says.

It says that ANY signal less than half the sampling rate can be represented. It's all been proved. Periodic has nothing to do with it.

A sine wave doesn't carry any information anyway. Seen one cycle, you've seen them all.

well i would dispute that, even if we are hearing it along with other things. i don't know what you mean by "the integration is part of everything from the "sigma" part of the delta/sigma converter to the actual physiology of your hearing. It's everywhere and inescapable".

It means that you are always sensing an average of your nerve firings.

if you are trying to say that we are all incapable of discerning the individual as well as the collective, then i think you are wrong. the level of consciousness at which the individual is being discerned is likely to differ from individual to individual.

Sometimes when you are in a dark room with your eyes closed, you might see a flash of light caused by a particle striking your retina that came from outer space
. It's a form of cosmic consciousness, you know.

but anyhow, the comment of yours that lupo was referencing was something different. you had said that the dac "just tries to keep up", which appears to me to mean that the dac does its best to process the samples that it is presented with. from my reading, it appears that some systems average a number of samples before output, and some do not.

I didn't mean there was some sort of problem with the system and the D/A is struggling to work, but that the filter required in all conversion strings all the samples back together into the same continuous waves that came in.

however i do understand that i discern signals within the noise, and that the presence of noise does not render signals imperceptible

True. And the random nature of noise allows, over time, a signal to be detected below the noise floor.

i once heard that a stitch in time saves nine.

After Ben Franklin was struck by lightning, he started speaking in incomprehensible maxims like 'a penny saved is a penny earned' when we all know a penny saved is worthless.

Hope this helps!


DC

[dcollins]

Originally Posted by karyoky****

hi,
i think we agree about most stuff, and, for what its worth, i am not bothered by any disagreement. it is not necessary for me to have total agreement with everyone on every little thing.

That's good. And there is plenty of room for disagreements. Sometimes mis-information, or opinions stated as facts tend to take off and grow a life of their own though................

some of the stuff is pretty common knowledge, i guess. i do notice that some people tend to try to "extend" what is established to support whatever their current argument / theory / business plan may require.

And when it comes to digital audio, this is approximately once a week.

no, i believe you are wrong there. and there are some additional implementation issues arising from the "problem" that music is not generally periodic in nature.

It works for any waveform. Any. Really.

What implementation issue do you mean?

a sine wave carries plenty of information. well, it carries at least enough information to be a sine wave. what do you have against sine waves? boooooooooop.

In theory "no" information and "no" bandwidth for the sine.

But I still like them.

i will say this, briefly: if the ear is doing its own averaging of, say natural sounds [in my view this would be in addition to perceiving the individual "firings"], then i think it is ideally better to let the ear do that itself, and not to have the digital system doing it "for" the ear in advance of the ear's oppertunity to hear, because i do not believe the perception would be identical in those two scenarios.

I do not understand this.

i am not sure what distinction, if any, you are trying to make. i don't have any issue with the idea of detecting signal below the noise floor, over time or otherwise.

Just that the ear works much like an FFT spectrum analyzer.

well i guess you would know about ben franklin being struck by lightning. before my time.

Yep, 'ol Ben and I were tight.

i know for sure that a stitch in time saves nine, though. and, to be honest, i think that a penny saved is worth at least a penny [some old pennies are actually worth much more]. and i once took a whole bunch of pennies, put them together, and bought several dollars worth of stuff. this technique works with nickles, dimes, and quarters as well.

They could have used this math in that "16 bits" thread.

i have another weird dither topic that i think i will start a thread on in a little while, or maybe in the thread that just appeared recently [oh dear].

The anticipation is palpable!


DC

[Nordenstam]

Hello!

Originally Posted by karyoky****

i don't know what to tell you other than that trying to manipulate the the frequency content, or other aspects, of music, in the hopes of making up for shortcomings, by trying to "take advantage" of what you may speculate or believe to be the way people hear or perceive is nothing but deception and a degradation of the music.

you seem to think that people [or some people] only hear continuous sound in somewhat larger "chunks" that they put together later on in their heads or something. so you figure that just playing the music in such "chunks" of sound in the first place is just as good as playing music as continuous sound. i believe that is erroneous on its face.

just because the human being may [or may not] tend to parse information in ways that you speculate about does not mean that your presentation of the information to the human being in such a way in the first instance is the same thing as, or that it will result in the same effect or interpretation as would, the presentation of the complete information in the first instance.

What I'm trying to get at is that time is an essential part in the sensory experiences we have. One last try: take noise as an example. The criteria for noise is flat fourier transform. If you apply that logic across some larger span of time, you'll find it quite easy to create a noise. If you try to make a very short noise you'll have to resort to special mathematics to still have a flat fourier transform across a short stretch of time. (it's the basis of acoustical diffusors, BTW) If you take such a small stretch of time that does have a flat fourier tranform and loop it to make it continuous, you'll have a repetetive pattern that the ear will latch on to and recognize as a pattern. The only way to create a noise that'll sound like a noise to the ear is to consider a large amount of time when creating that noise. That will create a sound that does NOT have a flat fourier transform when observed across a small stretch of time. No matter where you analyze it, you'll find small patterns and lumps of energy thrown here and there in various places in the spectra. It will not look like noise when observed on the small scale, yet it'll be noise when played back in a linear fashion and observed using the ears or or other measurement tools that takes some time into to the consideration.

That's just one example. That's the way a lot of sounds works. They're nothing by themselves, they absolutely need a stretch of time to be what they're intended to be. The very idea of a frequency is dependent on time.

Originally Posted by karyoky****

i'm starting to think that you may spend a lot of time dealing with artificial sounds ["ambient d j stuff" ?], and if you were more oriented toward actual recording of natural sounds you may have a different outlook.

Both. Abstract music encompasses both classical and modern stuff. The criteria for what works is pretty much the same in both genres. Some of my favourite records are naturalistic stereo recordings("audiophile" records).

Originally Posted by karyoky****

with all due respect, i don't think you are in a position to say "how sampling works". there are lots of different methods and implementation used by different manufacturers and designers.

to recap, you are talking about david collins' statement that dacs deal with samples on an individual basis. just understant that at some point each sample has to be dealt with, even in devices that do averaging.

judging from some of the papers and data sheets i have seen, some dacs do averaging and some don't.

It's the basic sampling theorem as used by any and all manufacturers. The impulse response picture posted above and the papers by Lavry touches the heart of the issue.

This was your response to the picture of an impulse response of a single sample:

Originally Posted by oky****

i am not totally sure what to make of the graphs and so forth. i think your example shows that ringing and smear is a problem, and it seems to me that it is only to be exacerbated by dither.

..

Originally Posted by oky****

i understand the "sampling theorum". it is a sample-by-sample system as well. you always seem to want to trivialize the importance of the individual samples.

The individual samples only represent the values at the sample points. The sample points are conceptual ideas of infinitely thin slices of time. The real signal that was sampled and that is going to reproduced resides just about everywhere but the sample points. The value of the signal inbetween the sample points is the sum of a large number of samples. Therefore - most everything that comes out of the DAC is the sum of a number of samples. The only thing that's being put out on a sample-by-sample basis is the value at the sample points, those conceptual points of infinitely short bursts of energy. If you could put aside your personal issues and read the Lavry papers on sampling, you'd get a much better understanding of how it really works. The papers are purely informational, you do not have to buy any of the products after reading them.


Quick primer of the sampling basic: a single impulse with a perfect brickwall lowpass filter gives the sin(x)/x function, also known as the sinc function. It looks like this:


Notice that it crosses the zero line at every Pi, just like the sinewave/circle does. Every sample in a digital system looks like that, with the sample value being the peak of the sinc and the sample clock comming at every Pi. Each sample contributes a certain value at the sample point and zero value at all other surrounding sample points. The land inbetween though, is a far different matter. It's made up by the contribution of many such samples. By summing the value of the leading and trailing edges of many such sincs, the value of the land inbetween the sample dots is restored to what it should be. That was my point back in that other post. Considering the value of the samples themselves only tells the conceptual instantaneous value at the sample clock, it doesn't say much about the real values contained in the reproduced waveform.

Therein also lies the answer to this:

Originally Posted by karyoky****

i believe the "implementation issues" have to do with, among other things, a need to incorporate a few milliseconds of previous samples into the reconstruction analysis in order to achieve a certain level of accuracy. i think you would have to include all samples back to infinity or something to really do [but i think that is managed by assuming them to be zero].

...
i seem to recall reading about there being problems with the beginning of musical sections, and so forth. i could try to find some of the cites later perhaps.

..

Originally Posted by oky****

as far as any request for a b comparisons or whatever, it is denied. i am not a test facility, and neither are you, to my knowledge. i am always a bit amused by guys who are posting pictures and graphs of this thing and that thing that they have come up with on their personal computers, as if any of that is supposed to prove something.

We're all test facilities in these days. Anyone can do advanced measurements using perfectly ordinary computers. Couple that with the ability to use a common logical basis (science) and anyone can create proofs using a personal computer. It's also possible to create logical proofs by using WORDS only.. That's the way it's typically done. You know, axioms and all that science ish. I don't think those words have to be typed on a special keyboard to be significant..

Originally Posted by oky****

i'm not sure what you are trying to find out about, but i think audio precision and prism makes test equipment for those interesting in scientific testing. i think bob katz said something about needing a bit scope or something like that to examine digital audio. whatever.

Bit scope is the least of my worries when it comes to measurements. Got a bunch of nice tools, do know how to use them and keep using them too. I don't sit around and wait for someone to publish a book with graphs made on their personal computer when I can do the exact same sort of testing myself.


Andreas