Prodif2496: Att. Jim Roseberry
Catena
Since I read in your past message that you found differences recording at
96k, I wanted to know why, as I knew it isn't becasue what you think. Now,
after reading your good article at www.prorec.com , I can say you why your
conclussions are not valid.
You compared a 96K recording converted to 44K1 with a direct 44K1 recording,
didn't you? Well, what you in fact have "almost" demonstrated, is that 96K
doesn't make any difference. Let me explain:
1) When you convert 96K to 44k1, you don't notice any difference, right? It
demonstrates that 96K doesn't improve sound.
If you find any difference (very improbable), the reason is that the 96K to
44K1 conversion introduces some distortion. Additionally, maybe the software
isn't filtering above 22k05 well enough.
To do a really accurate test, convert 96K to 48K, or 88K2 to 44K1 (to avoid
distortion in the conversion) using software with a high performance
antialiasing filter (to avoid aliasing). High performance here means, for
example: 0.1 dB max ripple in the passband (20Hz-20KHz), >= 120dB stopband
attenuation, constant phase (FIR filter with odd taps). That is, like the
filters found in top quality 24 bit ADCs. In this case, you'll verify that
96K sounds exactly the same as 48K. In fact, a 44K1 converter samples
typically at 128x44K1, and then converts the rate digitally using a high
performance antialiasing decimator, with characteristics as described.
Anyway, you might agree that even with the weird 96K to 44k1 conversion you
did, 96K doesn't improve sound.
2) When comparing the 44K1 converted file with the directly recorded 44K1
file, what are you comparing is not the difference between sample rates (as
are already the same), but the antialiasing characteristics of your
converters. From the results you got (difference appreciated), I can say
that the antialiasing performance of these converters doesn't seem as good
as you expected. If you want to measure objetively the antialiasing
performance, let me know and I'll give you a test guide (you need only a
signal generator capable of injecting freqs above Nyquist, quality isn't
important for this). It's easy.
Theory also serves to understand listening tests. Without theory, it's easy
to deduct wrong conclussions. Theory and experiences are both necessary and
complementary.
I hope you understood everything. You can also find some background in my
article at www.prorec.com (DSP theory). Don't hessitate to ask me if you
have further
doubts.
Best regards,
--
J.M.Catena
ad...@sesa.es
Joel Braverman
Joel
Jim Roseberry
Perhaps I didn't make myself clear.
I compared a 96k recording to the same recording at 44.1
There was a noticeable difference.
I also compared recordings (originally recorded at 96k) that were SR
converted down to 44.1k, to the same recordings made at 44.1. The files
that were SR converted from 96k sounded better.
With all due respect, my findings aren't 'wrong'. There is no right and
wrong when dealing with aesthetics. I'm simply saying that I can hear a
difference. That is the truth. My wife, guitar player in our band, and I
could pick the 96k track each and every time in blind listening tests. That
isn't coincidence.
As a designer... I'm sure you know all there is to know. And I'm sure you
focus on those details. But... I do hear a difference when tracking at 96k.
My ears aren't wrong... they are the bottom line.
Jim Roseberry
Studio Cat Software (Audio software/hardware sales)
1-888-873-8855
http://www.studiocat.com
Jack and Jill Recording Studio
Studio Cat Productions
j...@studiocat.com
Mo Weston
Well Mr. Catena's theories aint worth a SHIT when it comes to a listening
test!
We are lucky to be an associate company of the local National Panasonic
Techics agents here in Q8 and have just had a demo of the new DVD
recorded at 96 K ! the sound even after theorising the "dogs ears"
scenario is AWESOME!
I am getting on in years and have lost some of my top end freq's (limited
to about 18Khz!!!) BUT my old ears can still define the VAST differnce in
sound quality, definition and downright Kick Ass audio!
As Jim knows we also have been tracking at 96 K the differnce between 48
KHZ at 16 bits 48 Khz at 24 bits AND 96 khz at float is chalk and cheese!
Thereorising is to be respected, BUT we are dealing with one of the Human
senses when it comes to aural excitment so thereories aint with a dick
shit to me! My ears, like Jims, rule!
L8er
Mo Weston JHP Q8
Johnny Smooth
you apparently don't think have hearing) that give you 24/96 in the first place.
Catena has the benefit of seeing the issues from both sides, music
and engineering. Catena's point (and mine also) is that you already
benefit from higher sampling rates in the encoding (oversampling)
without having to waste disk space storing at 96K. Objectively, is it
possible that the differences you heard were due to something else?
Otherwise identical equipment? Blind testing? My doctor
tells me I hear better than small children, so someday I'll
take the test myself... :-)
John
Mo Weston wrote in message <360A47...@ncc.moc.kw>...
Lionel L. Dumond
Can't you be executed in Kuwait for cussing like this? ;)
All kidding aside, your point is well taken, Mo... but in all fairness to
Catena (and to Jim), the ONLY listening tests that can possibly be
considered valid are of the "double blind" variety, and performed in a very
controlled environment using scientific methods, where EVERY possible
variable can be totally accounted for.
I was recently talking to Bob L. about this. He's very involved in helping
to determine the new standards for DVD-Audio (he was also instrumental in
the negotiations for the original Redbook specification) and he was
describing the DVD listening tests that Sony has set up in their labs in
Japan. The test methodology and specs were painstakingly developed over
several MONTHS at a cost of tens of thousands of dollars... and just the
instructions on setting up the test takes up and entire book! Even so,
there have still been arguments among engineers about the methodology.
I seriously doubt that either you or Jim went to these lengths in your
tests. I am NOT saying that either one of you didn't "hear" what you
believe you heard -- far from it. What I AM saying is that neither
conclusion should be accepted as the true result of a scientifically valid
test, but as what they are -- one man's opinion.
Mo Weston <jin...@ncc.moc.kw> wrote in article
<360A47...@ncc.moc.kw>...
Catena
>considered valid are of the "double blind" variety, and performed in a very
>controlled environment using scientific methods, where EVERY possible
>variable can be totally accounted for.
Couldn't say it better.
--
J.M.Catena
ad...@sesa.es
Jim Morris
Here's something else to add fuel to the fire---
For many years now, some of the most sought after and respected recording consoles are the older 'discreet' NEVE models in the 80 series. Rupert Neve has stated that he feels that the reason for their popularity with a great many professionals is that his designs have flat frequency response all the way past 50kHz. A possible theory to explain this would be the interaction of high frequency signal components that create a 'sub harmonic' that IS in our hearing range. With 48kHz sample rates, no signal above 24kHz can be acurately reproduced. At the 96kHz rate we would still get components up to 48kHz. Mr. Catena has imparted to me his position about 'over-sampling' and his belief that current converter designs make the 96kHz sampling rate a "waste". I certainly have a great deal of respect for his opinions, as I feel all us Cakewalkers should, but I too believe that the difference is quite audible between 44.1 and 48. If that's the case, it would seem that 96 would be at least 'a little' better.
BTW -- Anybody going to AES this weekend? If hurricane Georges lets our flight get out of Tampa, I hope to see some of you there!! Mr. Catena, let me buy you a drink and get you to change my mind!!
Jim Morris
Morrisound Recording
Catena
Perhaps I didn't make it clear too.
In no way I was trying to say that your ears are wrong, and I agree that
listening is the bottom line with any audio comparision.
What I tried to explain is that your comparision is not valid because there
are other parameters related with the sample rate that you overlooked:
antialiasing filter at ADCs, during conversion, and conversion between non
integral multipliers. If you want to verify if high sample rates makes a
difference, you must keep these other parameters out of the equation as I
explained, what you aren't doing. A correct test methodology is required to
make the listening test valid, and your method is far from being correct,
because you are intruducing many factors in the comparision, not only the
sample rate. That is, you dind't isolated the cause of the difference you
hear.
I respect you, and appreciate your contributions, please don't take this as
offense. I'm simply trying to help to avoid a big mistake, as in this area I
know quite a bit.
Jim Roseberry
>a lot of unnecessary elements that are known to produce degradation (like
>96K to 44K1 conversion, to name just one of them).
The reason why I didn't SR convert to 48k is OBVIOUS! Ain't NO ONE gonna'
do that in a real recording/mixing/mastering environment. Folks are going
to want to know the results of converting down to 44.1k so as to burn Red
Book CDs. Isn't this obvious?
I compared recordings made at both 96k and 44.1k (without ANY SR
conversion)... and 3 separate people in my studio could pick the 96k track
every time. Irregarless of theory, irregardless of how many times you say
my findings are incorrect, the bottom line is the same. Sampling at 96k
produced BETTER results.
I also compared 44.1k recordings made by several other audio cards, to the
96k recordings thru the 2496s converters. Again... the difference was
clear. You could hear that the 96k recordings had a more natural open high
end.
Let me also say one last thing... My article says that the listening test
were NOT scientific measurements!
And let us get some perspective. Say I was reviewing a Neuman U87. My
conclusion is that it subjectively sounds georgeous when tracking in my
studio... Am I to believe that because I didn't take scientific
measurements (or track at Ocean Way) that my findings would be totally
invalad and false??? NOPE!!!
A subjective listening is exactly what it is... subjective. The is NO
absolute right or wrong. Period.
When all is said and done, I stand by my original article's contents. I
COULD INDEED hear the difference. I could indeed hear the difference. I
could INDEED hear the difference.
Craig
Jim's and other regulars opinions on this newsgroup but I feel I must throw my
two cents in so here goes:
As a budding Electronic Systems Engineer myself, I've come to realize that the
world, in fact the Universe is governed by CONCRETE laws of math and physics.
We live in this Universe and exist BECAUSE these laws are true and unwaverable.
The human ear ONLY responds to certain frequencies. The mathematical concepts
put forward by Fourier and others are quite concrete and sound. These
individuals had more insight about the workings of the Universe in their feces
then 99% of the human population has in their heads.
I read both articles at prorec.com and I have to believe the side of scientific
reason and not here say. I will however admit that POSSIBLY recording at 96khz
sample rate may sound better (I have not heard it myself) but TRUST ME if that
is the case, there IS a scientific reason for it. This is not a magical
world. We do not live in a Xanth novel (Piers Anthony) and taking LSD does not
give you more insight, it just gets you high (real high at that).
Craig :)
Catena
>do that in a real recording/mixing/mastering environment. Folks are going
>to want to know the results of converting down to 44.1k so as to burn Red
>Book CDs. Isn't this obvious?
NOT !!! Not if what you want to know is if 96K sounds better.
You might use 88K2 to 44K1 as well if you could.
If you convert 96K to 44K1, you're not evaluating only the sampling rate
effect, but also some other things that are known to produce degradation.
You knew it.
>I compared recordings made at both 96k and 44.1k (without ANY SR
>conversion)... and 3 separate people in my studio could pick the 96k track
>every time. Irregarless of theory, irregardless of how many times you say
>my findings are incorrect, the bottom line is the same. Sampling at 96k
>produced BETTER results.
No, because you must assume that your ADC has undetermined characteristics
at each sample rate. To keep it out from the equation, you must convert the
same data sampled at higher rate with a decimator whose antialiasing filter
characteristics are known to be better than noticeable.
>I also compared 44.1k recordings made by several other audio cards, to the
>96k recordings thru the 2496s converters. Again... the difference was
>clear. You could hear that the 96k recordings had a more natural open high
>end.
Again you can't know if the differences are because the ADCs' antialiasing
or the sample rate.
>A subjective listening is exactly what it is... subjective. The is NO
>absolute right or wrong. Period.
OK, then you stated an oppinion, not a fact. You still can't say that 96K
sampling rate improves sound quality, as you didn't tested this.
>I COULD INDEED hear the difference.
I heared it the first time !!! What I say you is that it isn't because the
sample rate !!! You are in fact finding differences between two 44K1 files
from the same material, each result of different conversions. Does it show
something to you?
---
And now, why don't you test again following the methodology I sent you to
verify if really 96K makes a difference?
And please, don't get angry. We are just friends discussing to find our way.
Jim Roseberry
>I read both articles at prorec.com and I have to believe the side of
scientific
>reason and not here say.
You call actually USING the converters 'here say'? Puhleeze!
(Lionel your royalty check is in the mail <g>)
Do you realize how rediculous that sounds to me when I'm the one currently
USING and HEARING the 2496s converters?
> I will however admit that POSSIBLY recording at 96khz
>sample rate may sound better (I have not heard it myself)
Possibly??? Maybe??? But then you admit to NEVER even hearing a 96k
recording.
>but TRUST ME if that
>is the case, there IS a scientific reason for it. This is not a magical
>world.
No kidding! My point is that the article was NOT about the WHY, but rather
the SOUND.
I'm not talking hypothesis here... I've been tracking and listening. That
my friend is HARDCORE evidence. Its down right laughable that folks who
aren't using the converters are trying to tell me what I'm hearing.
So... you contemplate your theories and make hypothesis.
I'll be tracking at 24Bits 96k.
SoundForge, Cakewalk PA 8.0, Samplitude 2496, SAW Pro ALL support up to 96k
Sample Rate. But I'm sure those companies were just wasting their time in
doing so...
Jim Roseberry
Recording with the 2496s converters at 96k WITHOUT QUESTION produced results
that were superior to recordings made at 44.1k (with other cards and using
the 2496s converters).
>You might use 88K2 to 44K1 as well if you could.
The 2496s doesn't support the 88.2k Sample Rate. Or I cerainly would have!
>If you convert 96K to 44K1, you're not evaluating only the sampling rate
>effect, but also some other things that are known to produce degradation.
>You knew it.
I know this. But when USING the 2496s converters, folks are going to want
to know what the bottom line product sounds like.
>I heared it the first time !!! What I say you is that it isn't because the
>sample rate !!! You are in fact finding differences between two 44K1 files
>from the same material, each result of different conversions. Does it show
>something to you?
Do you actually think that performing a SR conversion from 96k to 48k would
sound WORSE than converting to 44.1k? You know the answer to this.
A couple of weeks ago, I would have thought I was full of shit too.
Is it due to filter slope being more gentle when using 96k? Is it the
Sample Rate itself?
Either way, it simply SOUNDS better. So... to that end I again say,
"Recording with a Sample Rate of 96k produces results superior to Sampling
at 44.1k."
Dave Matthews
> The human ear ONLY responds to certain frequencies. The mathematical concepts
> put forward by Fourier and others are quite concrete and sound.
hold that thought about 'The human ear' <g>
I for one believe in the arguments for higher sample rates, I believe
that greater bit depth is more important initially...But I've read
enough to understand why such higher sample rates like 96k could make it
easier to sample sounds. I can't wait until I can try a test and hear
for myself...
And now for some really off the wall stuff:
I have no problem with the validity of the documented frequency response
data of human ->ear<- hearing. I am however, intriqued by ferrite
particles in bird's brains and the skulls and spines of other critters
that have been shown to permit the animals to navigate using the earth's
magnetic field. I personally take this and other biological magnetic
response phenomenons as evidence that there is more to our
human-to-nature interface than sight, sound, touch, etc. I've read a
few articles where people were exposed to real instrumental music
recorded at high bandwidths, played back in full bandwidth (important
point), and with the bandwidth limited to 20Hz. Reportedly the people
exposed liked the non-limited version better. (In the practical audio
field it is hard to find an accoustical transducer, especially in the
consumer car/boombox realm, with flat frequency response to 100k or so,
but of course that's not the point of the higher sampling rates in the
Prodif2496 discussion.) What I find fascinating about this is the idea
that perhaps we humans have other sensors to detect these higher
frequency sounds, in ways traditional hearing tests don't, uh, test
for... So in fact, if we are to trust what the 100kHz audio pundits
tell us, we need to sample at 200k or so, and our playback systems had
better be capable of full flat response up to and beyond 100kHz! Hee!
> This is not a magical
> world. We do not live in a Xanth novel (Piers Anthony
Frankly, and in all seriousness, I find this world to be very magical,
but not in the Xanth way...we are living in amazing times..
Dave
--
hear Ruby Amanfu and Adobe Pagoda at Lost Frogs Records:
http://www.lostfrogs.com
Donny H. Grace
> ..... folks are going to want to know what the bottom line product sounds
> like.
So if you test drive a Cadillac on a rocky road (or the notorious railroad
track) and test a pickup truck on the freeway, is the bottom line that the truck
rides better and possibly even sounds quieter? <g>
_____________________
Donny H. Grace
GraceFull Productions
Craig
> This is not a magical
> > world. We do not live in a Xanth novel (Piers Anthony
>
> Frankly, and in all seriousness, I find this world to be very magical,
> but not in the Xanth way...we are living in amazing times..
>
> Dave
>
>
Ah yes, amazing indeed.
The fundamental reason for these amazing times is that we 'stand on the shoulders
of giants'. These giants (engineers, mathematicians and scientists of the past and
present) all had a firm foothold in reality. Their methods and/or theories were
and are constantly bombarded by reason and fact. Those methods and theories become
laws and realities when the community of educated people can no longer come up with
any ammo to prove otherwise and accept the theory put forth.
I think it's fair of Catena, myself or anyone interested to enquire about methods
when encountering a blanket statement that seems to go against what we KNOW (not
saying that there isn't more to know) to be true. I'm not saying that the Jim is
wrong or that the theory of extra senses is wrong (seems like it's straight out of
the X-Files series) but I'm one of those kind of guys that NEEDS some sort of
proof. I'm sure that you can agree with me that even a ten year old can speculate
but that doesn't mean I'm inclined to believe him/her.
This seemingly magical world and amazing times were born out of CONCRETE reality
and physical laws of science and mathematics. Built by engineers and scientists.
That may be a stale unmagical truth but by God it is the truth.
Honestly, if I were to tell you the world was flat when you knew damn well that the
world was round you'd have a hard time believing me unless I somehow proved to you
that new evidense showed that all this time what we understood to be a sphere is
really a flat surface. This is more or less the same type of arguement and also
why it's hard to take at face value.
My opinion (for whatever it's worth)
Craig :)
Dave Matthews
>
> >So in fact, if we are to trust what the 100kHz audio pundits
> >tell us, we need to sample at 200k or so, and our playback systems had
> >better be capable of full flat response up to and beyond 100kHz! Hee!
>
> Components):
> - Freq response: 26 Hz - 100 KHz +- 4dB, 33Hz-20KHz +- 2dB
> - Phase shift: 5KHz-100KHz +- 3 degrees
> - Ionic tweeter, of course.
That would do it!
> Then, you *only* need ears capable of hearing up to 100 KHz... As Bruce
> said, at least your dog will appreciate it for sure! (-;
The point that some make is that your ears would not be the audio
interface used to perceive that frequency sound! That ferrite on the
brain stuff again...or some other vibration detector sensitive to those
frequencies.
Dave Matthews
> The fundamental reason for these amazing times is that we 'stand on the shoulders
> of giants'. These giants (engineers, mathematicians and scientists of the past and
> present) all had a firm foothold in reality.
Absolutely
> wrong or that the theory of extra senses is wrong (seems like it's straight out of
> the X-Files series) but I'm one of those kind of guys that NEEDS some sort of
> proof. I'm sure that you can agree with me that even a ten year old can speculate
> but that doesn't mean I'm inclined to believe him/her.
The fun thing about science is that new discoveries are constantly
occuring...mind you I'm not suggesting that we *can* hear those
frequencies, but I'm not willing to discount the possibility. We get to
research our speculations...
> This seemingly magical world and amazing times were born out of CONCRETE reality
> and physical laws of science and mathematics. Built by engineers and scientists.
And our understanding of this physical world is constantly improving and
increasing, even altering previously held notions of the physical
laws...
> My opinion (for whatever it's worth)
worth a lot!
Gary Edelman
This thread has become quite informative & amusing but it sounds like
you & Jim are not talking the same language. He's talking about what his
ears hear & you're talking statistics. My guess is that you're probably
both right but thats not my question. What I'm trying to understand is
are you saying that all things being equal a 96k recording will not be
superior in quality to one recorded at 44k or that when a 96k recording
is converted to 44k the results would be no better than if it was
originaly recorded at 44k?
gary
Catena
>tell us, we need to sample at 200k or so, and our playback systems had
>better be capable of full flat response up to and beyond 100kHz! Hee!
You might be interested in the Monitor One, from AEC (Audio Engineering
Components):
- Freq response: 26 Hz - 100 KHz +- 4dB, 33Hz-20KHz +- 2dB
- Phase shift: 5KHz-100KHz +- 3 degrees
- Ionic tweeter, of course.
This is the toy I used to evaluate the perceivable bandwidth working for a
pro audio manufacturer about 14 years ago.
The only disadvantages are its price (Oooohhhhh), and the ozone generated
(puuffff).
Then, you *only* need ears capable of hearing up to 100 KHz... As Bruce
said, at least your dog will appreciate it for sure! (-;
--
J.M.Catena
ad...@sesa.es
Hugo Martinez
> At the 96kHz rate we would still get components up to 48kHz.
Speakers produce harmonic distortion. When compared to amps, it's a
lot (could be why speaker harmonic distortion specs are seldom if ever
quoted).
I would not be surprised if a 96k sampling rate would sound "better"
to me or at least different. Maybe it's not because I'm hearing
higher frequencies, but because I'm hearing the harmonic by-products
of those higher frequencies.
Just a thought.
Catena
>are you saying that all things being equal a 96k recording will not be
>superior in quality to one recorded at 44k or that when a 96k recording
>is converted to 44k the results would be no better than if it was
>originaly recorded at 44k
Yes to both cases, the result will be the same if the following conditions
are met:
1) The 44K1 ADC must be something like a sd x128 oversampling with a good
decimator: 0.1 dB ripple in the pass band, and better than 100 dB
attenuation in the stopband beginning at 22k05. These ADC chips are recently
available and are still a bit expensive (about $40 per piece), and thus
aren't used in all gear we would like.
2) The analog antialias filter must be something like: passband up to 20 KHz
+0-0.5 dB, attenuation above at 5.6 MHz (128x44k) better than 100 dB. As
this filter has 128 octaves for the transition band, so it can be done
easily and without phase shift in the passband (up to 20 KHz).
3)Once recorded, to hear the same a 44K1 stream as a 96 K stream, the 44K1
DAC must have flat freq response up to 20 KHz with less than 0.5 dB ripple.
Note that what a sd x 128 ADC does, is sampling at 5.6 MHz (WOW!). After
signal is digitized, it's converted (decimated 1/128) to 44K1, but inside
the converter chip. So, sampling at higher rates to later convert to 44K1
makes no sense.
The only possible differences are in the quality of the rate conversions. As
the hardware can now do it as well as the software, let it make the hard
work and save resources...
--
J.M.Catena
ad...@sesa.es
Catena
>interface used to perceive that frequency sound! That ferrite on the
>brain stuff again...or some other vibration detector sensitive to those
>frequencies.
ROFL. If we don't have this ability (what seems at least probable as
experiments revealed), then there would be a way:
Lately, some scientists have been experimenting injecting signals form
artificial sensors directly in brains. Thanks to that, even blind people can
see something now, injecting signals from a camera in their optical nerves.
It isn't fiction. So, there is a possibility that extended frequency
response microphones might inject a treated signal to our hearing nerves,
and then, we might hear up to 100 KHz. When most people would have this
"installed", then we'll need to sample above 200 KHz. I don't know if we
would like the music with 100KHz bandwidth, all what I can say is that I
like it a lot "limited" to 20 KHz.
Enjoy!
--
J.M.Catena
ad...@sesa.es
A.J.
says...
Interesting thought. What about some sort of subtle IM-like distortion(in
the analog domain) between the frequencies we can hear and the ones we
can't? Also could the high frequencies be altering the response/linearity
of an analog gain stage - sort of like how ultra-high-frequency parasitic
oscillation can 'shut-down' a tube amp?
Pdemotech (Pete Leoni)
Khz, what he is hearing is the benefit of better filtering, Better, because
it has been easier and cheaper up to now to properly filter *higher* sample
rates than it is *lower* sample rates, Therefore technically, what we
really need is proper filtering. Proper filtering would eliminate the need
to sample at rates greater than the Nyquist theorem dictates while reducing
processor overhead. Practically however, because of current "less than
optimum" and economically oriented filter design, we never get to hear the
optimum quality that 44.1 Khz sampling is capable of achieving. Jim may
very well be hearing an improvement in quality, albeit at the expense of
unnecessarily bloated processor overhead. Kind'a like swatting a fly with a
Louisville Slugger, It gets the job done though.
pete
Catena <ad...@sesa.es> wrote in article <6ufjf4$5...@hope.harvard.net>...
> >What I'm trying to understand is
> >are you saying that all things being equal a 96k recording will not be
> >superior in quality to one recorded at 44k or that when a 96k recording
> >is converted to 44k the results would be no better than if it was
> >originaly recorded at 44k
>
John Harragin
theorem. 2. A technique where each sample from the data converter is
sampled more than once, i.e., oversampled. This multiplication of
samples permits digital filtering of the signal, thus reducing the need
for sharp analog filters to control aliasing.
from Rane Professional Audio Reference
Oversampling already exists - and it provides the benefits that the 96k
advocates claim - without cutting our machines ability in half! There
are other potential problems with mixing in 96k. Along the lines of us
torturing our pets. For instance unintentional high frequency signal
can go undetected and be sent to monitors not designed to handle these
frequencies. I don't know how serious this could be but I'm sure it can
cause troubles.
Anyway, I think the people citing better sound, shimmer... Are probably
not compairing their 96k cards to real high quality oversampling cards.
In the case of the same card that record at at 96k or 48k (without over-
sampling) that comparison is not being made.
John
"Pdemotech (Pete Leoni)" <demo...@datasync.com> writes: > So, the bottom line is....... if Jim is hearing an improvement @ 96
Joel Braverman
"hey mom, I want to be an engineer"
"Sorry kid, I just can't afford the implant operation"
"but mom, my sister got _her_ implants
"That's different - your talking about making money in the future, your
sister is making money right now!!"
Chris Townsend
<6ue76c$a...@hope.harvard.net>...
> Either way, it simply SOUNDS better. So... to that end I again say,
> "Recording with a Sample Rate of 96k produces results superior to
Sampling
> at 44.1k."
>
You are making quite a bold claim, that recording at 96K simply sounds
better than that sampled at 44.1K. Your Pro Rec article also seemed to
make the same conclusion, but I think all you can accurately claim is that
in your opinion the Sek'd 2496s converters sound better at 96K that at
44.1K. But this does not show that the difference you hear is due to the
increased sample rate, or any other effect directly related to the
increased sample rate (like filter slope). Although it certainly may be
the case that 96k recordings generally sound better, there are numerous
reasons that this might not be the case even though, you "hear a
difference" when using the Sek'D converters.
One possible reason for the difference you hear is that any set of speakers
and amplifiers will produce inter-modulation distortion to some degree.
When presented with ultra high frequencies this inter-modulation distortion
will likely have components within the audible frequency range, and this
may contribute to the difference that you hear. Now of course this
wouldn't negate that fact that some people prefer the 96k recordings, but
it is important to note the reason they prefer the 96k recordings is really
because they like the subtle high end that the inter-modulation distortion
adds. And don't think that this is a far fetched scenario, since even the
best speakers in the world have considerable harmonic and inter-modulation
distortion (1% range), which could easily contribute to the difference that
you hear.
Another reason is that there could be subtle differences in the converter
design, when running at 44.1kHz which does not show up a 96kHz, yet are not
directly related to the sample rate. If this is case then your conclusion
that the Sek'D converter sounds better at 96k is perfectly valid, but your
conclusion the 96k recording in general sound better would most likely be
false. A some what likely mannifestation of this is that the converters
may have a slightly different output level when using different sample
rates. Of course this difference would be very small, on the order of less
than a tenth dB, and would be difficult to measure accurately with standard
studio equipment. But subtle differences like these can certainly change
the subjective impression of the sound.
I guess the conclusion here is that you have given very little evidence,
subjective or objective, that 96k recordings in general sound better than
44.1k. Of course none of this negates your opinions regarding the 2496s
sounding better when sampling at 96K, but your overall claim seem to go
well beyond just that.
Chris
--------------------------------
Chris Townsend - DSP Engineer
Arboretum Systems, Inc.
http://www.arboretum.com
--------------------------------
Jim Roseberry
You have wonderful hypothesis... and have many opinions.
Have YOU varified ANY of your hypothesis before claiming I'm full of shit?
I mean... with REAL working converters??? If so, which ones?
If not, here is your no-risk chance. <g>
Pick up a set of the 2496s converters from me... and if you don't agree with
what I've said, I guarantee your money back. I can offer NO better proof
than this.
There is NOTHING I'm going to say that will change your mind...
And by the same token, words typed here aren't going to change what I hear
when recording at 96k. And I do hear a difference... and so do others...
Mo Weston runs a HUGE facility is Q8. He is actually USING 96k audio... and
he also hears a difference. Now, that's just a little odd, the guys who USE
the tool make observations, and the guys NOT using the tool try to dispute
what we hear.
Ok Ok... I'm totally FULL OF SHIT! Happy??? But we'll see the truth
revealed over time.
When you actually have owned and USED a set of 96k converters for a coupla'
months, give me a call and we'll discuss who's full of crap and who was
right. Fair enough?
Cakewalk PA 8.0, Sound Forge (the first to support 96k), Samplitude 2496,
SAW Pro... ah what the hell, the developers had nothing better to do...
Mo Weston
>
> Have any of you actually USED a set of 96k converters?
> You have wonderful hypothesis... and have many opinions.
> Mo Weston runs a HUGE facility is Q8. He is actually USING 96k audio... and
> he also hears a difference. Now, that's just a little odd, the guys who USE
> the tool make observations, and the guys NOT using the tool try to dispute
> what we hear.
OK here's one for you all to opinionate,hypothosize and therorize.
I tracked the latest Sony Seafront Jingle here at 44, 48 AND 96 Khz,
dropped all three to a demo Sony EF60 Type I cassette and asked the
Ad Agency (Fortune Promo Seven) and the Client (Sony) to pick out the
demo they thought sounded the best. EVERYONE bar none picked out the 96K
recording. Now if one of my top clients picks out the 96 Khz tracked demo
dropped to a shitty oxide cassette then something tells me END OF DEBATE!!!
Mo Weston JHP Q8
Mo Weston
Yo Jose,
Your therories are sound enough, BUT have you thought that other senses come into
the playfield and add to the listening experience.
It's a known, tested and proven theory (I ran a hearing test recently) that my old
ears top out at 18.7 Khz! (boo hooo).and the range at the bottom end down to about
50.
It's also a proven fact to me that freq's below 20 Hz can be FELT rather than heard,
could it be that the same can be said of the top end????
I've long given up theorizing why higher sampling freq's, even above the Human hearing
limits simply SOUND BETTER....
CAP THAT!!
Mo Weston JHP Q8
Gary Edelman
> conditions
> are met:
I appreciate your answer & I respect & admire your technical knowledge
but my logic tells me otherwise. I would not dare argue facts &
statistics with you so I only pose a logical question.
If in midi you use 960 ticks per quarter note as opposed to 480 ticks it
would be impossible for a listener to tell unless they were totally
familiar with the peice & maybe possibly feel a slight difference. In
sampling with more bytes sampled per second it's only logical to suggest
that it helps to get closer to the natural sound of what is being
recorded whether or not certain frequencies are heard by the human ear.
I also believe that there are subtle overtones & harmonics in music that
can penetrate the human body beyond the ear. But since we are dealing in
numbers isn't the smaller denominations always more exact. Though .67 is
the equivilant of 2/3, isn't .667 more exact & .6666667 even more so
though none are truly exact. If you had said only that converting from
96 to 44 was no better quality I would understand. But in saying that 96
is no better than 44 it just doesn't compute in my way of thinking.
We'll probably be recording 192K at 64 bits in the future, who knows.
Gary Edelman
Hugo Martinez
>Now if one of my top clients picks out the 96 Khz tracked demo
>dropped to a shitty oxide cassette then something tells me END OF DEBATE!!!
You bet. I'm going tomorrow to trade-in my DAT deck for a couple of
cases of Type I cassettes. Then I'm gonna dump my PC for a vintage
Portastudio.
George Steber
different perpective. I have 2 digital oscilloscopes in my lab, both 8
bit. One is a Heath that has a max SR of 100KS/s, the other a Tek 360S
that goes to 1 Gs/s (thats 1000 MS/s). So now I sample a sine wave of
frequency 50Khz. What happens? Well the Heath (at 100Ks/s) obtains a
rough approximation of the signal. It is trying to reconstruct the sine
wave with only 2 samples per cycle. Its an older model and does a poor
job of sampling and reconstruction. But its ok for lower frequency audio
signals. Now sample with the Tek at 200 Ks/s. Wow. A big difference. The
sine wave is reconstructed almost perfectly. Is this due to the higher
sampling rate alone? Lets see. Set the Tek to sample at 100 Ks/s and
display the waveform. Amazing. It is still very good. Why? Because the
Tek has a better a/d, better s/h, better anti-aliasing filters, better
reconstruction filters, and better d/a. So we see that while SR is
important, it is not the only factor in good digital signal processing.
George
Johnny Smooth
>numbers isn't the smaller denominations always more exact. Though .67 is
>the equivilant of 2/3, isn't .667 more exact & .6666667 even more so
>though none are truly exact.
True, but you're suggesting that there is more information to be conveyed
in the extra bits.
The 128x converters you use now really sample at around 5.6MSamples/sec.
That's definitely better than 96K. The reason for the overkill is to push
the "aliased" copy of the frequency spectrum as far away from the original
as possible so that you can use a nice linear FIR filter to remove the aliased
copy. Once that is done, what is contained in the data is only the info
below 22K (the rest got filtered out). Well, this info can be exactly
represented
at 44.1. Storing it at 96K is like writing a letter on a piece of paper that is
twice as big as it needs to be. No more info, just a waste of space.
Now some caveats. If you're arguing that we percieve higher than 22K,
then even the theory says that gets thrown away, so you would notice
a difference. Another thing that
might make a difference is that outputting the data at 96K does give
more room for the FIR filter to work, but that one (for me) I find hard to
believe would make that much difference considering the room that
is already there due to oversampling. You could use the same filter if
you oversampled at 256x and still store at 44.1 (I don't know if any
current ADC's work that fast though). Does anyone (Jim?) know what
the current batch of 96K converters sample at? If they are still sampling
at around 5.6 M/s (or multiple of 96K), then the only difference is the FIR
filter.
For my money, I'd rather get more tracks, bits, and real-time effects than
worry about what I'm missing between 22K and 48K (since all will be competing
for time on my Pro200). My speakers don't go that high, and I'm guessing my
ears don't either.
John
Gary Edelman
conveyed in the extra bits.
> It seems logical to me. I would think that oversampling only helps to
> prove my point. If more bitst helps for A/D conversion than why not
> more bits of information stored.
gary
Catena
times here. But many people can't understand yet, or don't want. That was
the reason to write my articles about DPS theory for www.prorec.com (online
now), trying to offer a comprehensive path from scratch. I hope that after
the first three or four articles, the discussion about this and other
related issues, if still exist, would be at least based in serious
arguements, becoming a constructive discussion and not a meeting of sheeps
(no offense intended).
Best regards,
--
J.M.Catena
ad...@sesa.es
Johnny Smooth escribió en mensaje <6uj6f8$7...@hope.harvard.net>...
>Gary Edelman wrote in message <360C9E9D...@gte.net>...
>>numbers isn't the smaller denominations always more exact. Though .67 is
>>the equivilant of 2/3, isn't .667 more exact & .6666667 even more so
>>though none are truly exact.
>
>True, but you're suggesting that there is more information to be conveyed
>in the extra bits.
>
Johnny Smooth
>True, but you're suggesting that there is more information to be
>conveyed in the extra bits.
>
>> prove my point. If more bitst helps for A/D conversion than why not
>> more bits of information stored.
>
>gary
The purpose in oversampling in the A/D isn't to get more information,
it's to make life easier when the aliased copies are being removed
(and to avoid prefiltering in order to bandlimit the signal coming in).
If we didn't oversample, then the filter would have to roll off from
passing the signal to cutoff in only 2.05KHz (22.05K - 20K). That's
because the copy of the signal starts at 1/2 the sampling frequency
( 44.1K / 2). However, if we sampled at 5.6 Megahertz, then
the aliased copy doesn't show up until 2.8 MHz. That's a lot more
room to design a much better filter to remove the alias copy. At this
point we could use the signal at 5.6MHz, but there's no need to keep
it at that high a sampling rate - the *same* information can be represented
at a much lower rate, as long as the rate you're downconverting to is
higher than the cutoff of the filter that you just used. Honest. That's
Nyquist talking, not Catena or me. :-) So the point is that *if* you
filtered at 22K, then you gain nothing by staying at 5.6M, or even stopping
at 96K; you lose nothing by going all the way down to 44.1. As I have
always added though, if the debate is whether we can perceive above
22K, then I think the jury is still out.
I'm not suggesting that people aren't hearing differences at 96K; experience
does count for a lot. But to point to one variable in a multivariable
experiment
(96K in this case) and claim it's the cause is unsupported without further
testing. I expect one rebuttal is "Who cares, it sounds better!". My
response is that 96K is by no means "free". Do you want to go back to
getting (only) 12 tracks and maybe one real-time effect, and all the while
*doubling* your disk requirements just for that little
extra that you "might" hear, in spite of the fact that your monitors only
go up to 20K and your ears probably less than that? That's fine if you
do; it might be worth it in your situation. Just means I'll be paying less for
the 44.1K/48K converter I'll be buying soon. :-)
Hope this was helpful,
John
Pdemotech (Pete Leoni)
information in more bits, this is fundamental, dynamic bandwidth as opposed
to frequency bandwidth.
pete
Gary Edelman <gar...@gte.net> wrote in article
Andrew Lovell
>
> Have any of you actually USED a set of 96k converters?
> the tool make observations, and the guys NOT using the tool try to dispute
> what we hear.
>
> revealed over time.
I trust listening tests much more than theora, graphs and scope
traces. So far, the Catena side of this dispute has not mentioned any
music they recorded through different convertors at such and such
sampling frequency, I would guess because they haven't done so. This is
much akin to those philosophers of ancient times who would argue
endlessly about the number of teeth in a horses' mouth without going
down to the stable and opening up the appropriate equine orifice.
Please correct me if I'm wrong about your lack of experience, I don't
have time to live my life in this ng the way some do. Gentlemen, until
you've done the listening tests, you haven't contributed anything to
this discussion. I'm not interested in a dissertation on digital audio
or parts cost. Only the bottom line counts, and that's sound quality.
If lay people are able to pick this stuff out on workaday studio
speakers, maybe there's something going on that you need to get hip to.
If perhaps your theories need expanding or modification, I'm sure you
can reverse engineer the reasons for the increase in sonic
wonderfulness. It's been done before.
Yours truly,
Andrew Lovell
John M.
=======================>
> much akin to those philosophers of ancient times who would argue
> endlessly about the number of teeth in a horses' mouth without going
> down to the stable and opening up the appropriate equine orifice.
hehe equine orifice? I love it!
That would win any "name the band" contest. <g>
Johnny Smooth
for theories are usually those that understand them the least.
When a dozen people count the horse's teeth and
give you 5 different answers, how much faith do you
put in experience?
Good luck with your articles, Catena.
John
Tom Greb
You guys just keep the free theory lessons coming.
Most of us are big enough to make up our own minds
concerning what info is most useful to us.
Thanks,
Tom
Catena
>much akin to those philosophers of ancient times who would argue
>endlessly about the number of teeth in a horses' mouth without going
>down to the stable and opening up the appropriate equine orifice.
Stop here! What I'm explaining aren't *my* hypothesis. What I'm explaining
is what science has demonstrated mathematically and experimentally following
scientist methods, what you can read in all science books, what you can
learn in the university , what you can verify if you have the knowledge to
isolate the variables in an experiment, what all engineers are employing in
practice to develop all digital audio pruducts (and others).
Do you really think that scientists and engineers doesn't do listening
test? They are the ONLY professionals that knows how to do accurate, error
free experiments, including listening tests, of course. What I explained has
been demonstrated both mathematically and experimentally by scientists and
engineers, and verified along many years of developments. At the other side,
NOBODY has demonstrated that any point of this proven theory is wrong, only
statements like "I found a difference, and that's what matter" after a
wrong, uncontrolled, subjetive, non scientific experiment.
And don't come saying that science was wrong when said that the earth was
plain (that was not the science like we know today). We know today that the
earth is a sphere, and nobody can demonstrate that this is wrong, we have
seen that it's a sphere. The same way, exactly the same way, nobody can
demonstrate that Nyquist or Fourier were wrong, because it's DEMONSTRATED by
science (what inludes every kind of experiments) to be right.
In the next two chapters of the DSP theory articles for www.prorec.com (read
DSP science, not my hypothesis), I'll explain some issues that will help to
understand better this issue of the sample rates.
>Please correct me if I'm wrong about your lack of experience, I don't
>have time to live my life in this ng the way some do. Gentlemen, until
>you've done the listening tests, you haven't contributed anything to
>this discussion.
My "lack of experience" is because I studied both electronics and computing
engineering while working for a professional audio manufacturer about 15
years ago. Since then, my job has been always in the research and
development area, learning "a bit" about these issues. As I love music
making, what has been my favourite hobby since I was a boy, I always applied
all technology I learnt to the music production. I built my first home made
audio converters for my computer many years before the PC, and I had my
first sequencer and sound module many years before the MIDI . And I still
found time to record bands in studios and mixing in live in halls filled by
tens of thousands people. About seven years ago I was developing audio DSP
products, including rate converters, and evaluating its performance
(obviously including objetive listening tests as one of the many factors
involved).
You need at least 10 years of full dedication to match my "lack of
experience", and that's ONLY if you can learn from people like me. If not,
you'll waste these years inventing a wheel that isn't round.
Do you want to play hardball with me? Sure? (-;
--
J.M.Catena
ad...@sesa.es
Mo Weston
Dont trade in yer DAT deck just get a 2496 converter and drop 96 K to
your DAT.
I was merely pointing out that EVEN the 96 khz audio dropped to a Type I
Demo cassette was chosen as having a better sound and defintion.
Can you even begin to imagine how it sounds on DAT.
I'd like to think so !!!!!
Mo Weston Jhp Q8
Jim Roseberry
You certainly know the scientific side. I don't begin to dispute that.
But... I do hear a difference when working with 96k audio. So do many
other people. Read this month's EM. There you see listed the EXACT
findings that I heard in my 'full of shit' 'absolutely no credible data'
subjective listenings. Clearer more detailed high end, and a more 3D sound
stage... Sound familiar???????
What is so damn threatening about 96k audio?
As a scientist, how can you argue that a recording device has too much
resolution?
That is a theoretical impossibility. To assume that all sound above 20k is
completely insignificant is really quite arrogant. I'll use a short delay
(.1ms for instance) as an example. No one can 'hear' that delay, but it can
certainly make a difference to the stereo image.
The storage space? Big deal! Hard Drives are rediculously inexpensive
compared to just a couple of years ago, and they sustain about double or
triple the load.
Obviously, if you have a large production to track/mix 96k audio is probably
out of the question. But... if you are tracking singer/songwriters, trios,
etc., 96k recording sounds wonderful! I (without question) benefit from
recording smaller productions @ 96k. No matter what you or anyone else
says, my EARS hear the difference. That is the bottom line. And nothing
said changes that experience.
Watch the industry... pay special attention to what the large tracking and
mastering studios (the big boys with unlimited budgets) are using. I can
tell you right now that they are 'movin' on up.' Cue the Jeffersons theme
<g>...
Jim Roseberry
But... If you are interested in 96k audio, forget all this bo-jive and go
listen to a set of 96k converters.
Here is another perspective.
Instead of wasting time telling me I'm just plain wrong, why not try to
theorize and hypothesize why I'm hearing a difference when recording at 96k?
Also of note:
There are three different articles that mention 96k audio in this month's EM
magazine.
1. Mastering engineer uses a 24Bit 96k digital console that he absolutely
LOVES.
2. George Duke was going to mix at 96k, but the converters he wanted
weren't available.
3. I believe the other article (refering to 96k audio) was just a general
discussion that mentioned listening tests, amd the fact that people ARE
hearing a difference. These people site a clearer more detailed high end,
and a more 3D soundstage. Now where have I heard that before???...
Phat Bass
Jim Roseberry wrote in message <6uninl$l...@hope.harvard.net>...
>What is so damn threatening about 96k audio?
>As a scientist, how can you argue that a recording device has too much
>resolution?
Jim,
What Catena and others have tried to do is give a scientific reason for WHY
"it sounds better". As they said, perhaps the speakers are throwing
harmonic distortion which may be coloring the sound to be subjectively
nicer. You keep telling him to "use" something at that resolution, but what
are you going to do when he does and still has the same scientific theory to
stand on? I don't think either of you have run enough actual tests to have
formed a complete conclusion. Rather, I think you are speaking about the
SAME thing, but coming from different directions.
P. Bass
Jim Roseberry
>harmonic distortion which may be coloring the sound to be subjectively
>nicer
With all due respect... I'm not hearing harmonic distortion.
You are going to have to do better than that to explain why the high end is
more clear & detailed, and the soundstage is more 3D.
Again... I urge you to have a listen to a 96k set of converters. If more
folks would take the time to listen to 96k audio, these discussions would be
a LOT more productive.
Johnny Smooth
>Nothing wrong with theory...
>
>But... If you are interested in 96k audio, forget all this bo-jive and go
>listen to a set of 96k converters.
>
>Here is another perspective.
>Instead of wasting time telling me I'm just plain wrong, why not try to
>theorize and hypothesize why I'm hearing a difference when recording at 96k?
I thought that's what I was doing and not trying to tell you you didn't hear
a difference. The question is why. With speakers that roll off at 20K, how
can one perceive the extra bandwidth?
The thing to "fear" about 96K ( :-) ) is that it's a tradeoff between cost of
extra hardware and diskspace versus improved sound quality. Theory
says the improvement of 96K can come from one of two places: higher
frequencies represented (which I question), and a wider bandwidth available
for possibly a better anit-aliaising filter. The second could be achieved by
increasing the oversampling rate and then still storing at only 44.1K. If
someone could offer you the same sound quality that you're hearing with
the 96K converters with the extra benefit that you use 1/2 the disk space
and get 2x the tracks, wouldn't you want that? Like I said, I'm interested in
why you hear a difference, not trying to say you don't. If the "why" is due
to something other than 96K, then we all benefit by being able to work with
1/2 the amount of data.
>Also of note:
>There are three different articles that mention 96k audio in this month's EM
>magazine.
>1. Mastering engineer uses a 24Bit 96k digital console that he absolutely
>LOVES.
>2. George Duke was going to mix at 96k, but the converters he wanted
>weren't available.
>3. I believe the other article (refering to 96k audio) was just a general
>discussion that mentioned listening tests, amd the fact that people ARE
>hearing a difference. These people site a clearer more detailed high end,
>and a more 3D soundstage. Now where have I heard that before???...
I'll read (or re-read) the articles.
John
Phat Bass
Jim Roseberry wrote in message <6uo9if$4...@hope.harvard.net>...
>Again... I urge you to have a listen to a 96k set of converters. If more
>folks would take the time to listen to 96k audio, these discussions would
be
>a LOT more productive.
Sure, donate me some, and I'd be glad to :) Otherwise, I'll make due with
my ISA 16bit DMAN.
P. Bass
Jim Roseberry
>The thing to "fear" about 96K ( :-) ) is that it's a tradeoff between
cost of
>extra hardware and diskspace versus improved sound quality.
I realize this... I don't always track at 96k. For larger productions it
is out of the question.
>Theory
>says the improvement of 96K can come from one of two places: higher
>frequencies represented (which I question), and a wider bandwidth available
>for possibly a better anit-aliaising filter. The second could be achieved
by
>increasing the oversampling rate and then still storing at only 44.1K. If
>someone could offer you the same sound quality that you're hearing with
>the 96K converters with the extra benefit that you use 1/2 the disk space
>and get 2x the tracks, wouldn't you want that? Like I said, I'm interested
in
>why you hear a difference, not trying to say you don't. If the "why" is
due
>to something other than 96K, then we all benefit by being able to work with
>1/2 the amount of data.
I understand your point.
Randy Hammon
the 96K converters that sek'd have in their booth. Their new AD/DA converter
is absolutely wonderful. Rainer, the 2496 product manager from germany is
there and I had a nice chat with him. He showed the converter to me and I
could've sworn it had a setting for 88.2, meaning samp 2496 should do it now
or very soon.
Thanks,
-randy
Jim Roseberry wrote in message <6uo9if$4...@hope.harvard.net>...
>Again... I urge you to have a listen to a 96k set of converters. If more
>folks would take the time to listen to 96k audio, these discussions would
be
>a LOT more productive.
>
>
Jim Roseberry
The 2496DSP DOES support the 88.2k sample rate.
It will also allow you to capture 24Bit 96k audio to a standard DAT deck.
That's pretty nice for folks who do a lot of stereo live to DAT recording.
World-class... but they aren't cheap
Lionel L. Dumond
I am assuming that it must be like the new 24-bit Tascam DAT deck, which
will allow 24 bit recording up to 1/2 the rated time of the DAT -- in other
words, 45 mins on a 90 min tape...
--
Lionel L. Dumond
Producer / Senior Engineer
MusicMedia Productions
Portland, ME
Jim Roseberry <jimros...@sprynet.com> wrote in article
<6uoms5$b...@hope.harvard.net>...
Jim Roseberry
>I am assuming that it must be like the new 24-bit Tascam DAT deck, which
>will allow 24 bit recording up to 1/2 the rated time of the DAT -- in other
>words, 45 mins on a 90 min tape...
You are probably right... I'll find out and let you know.
Jim Roseberry
Studio Cat Software
Jim Roseberry
That was very helpful indeed!
Perhaps a converter 'shoot-out' would make a good prorec article.
Catena
>folks would take the time to listen to 96k audio, these discussions would
>be a LOT more productive.
Jim, this discussion is being productive, I think. I hope you find it
interesting too.
I can add some new information to better understand this issue.
There are reasons, supported by theory, that can explain your experiences.
These are:
1) 96K stores twice the bandwidth than 48K. Although the additional
bandwidth can't be perceived, it improves by 6 dB the dynamic range in half
the bandwidth (the audible band). This is a kind of dithering that I'll
explain in the next chapter at ProRec dedicated to conversions. With
converters with better than 120 dB dynamic range, the difference should not
be perceived because it exceed our hearing capabilities. But if the
converters are below that, the improvement is noticeable. I think Sekd
converters are rated at 113 dB(a), or about 106 dB unweighted. This explains
a part of the improvement you found. Obviously, as there are converters with
better than 120 dB effective dynamic range, the same (and even better)
quality can be achieved without resorting to higher sample rates.
2) It's sure that aliasing played a very important role in your tests. You
know already that an oversampling converter samples at a much higher rate
that the final one. And you found differences between two 44K1 files both
coming from a very high original sample rate, one from a direct decimation
in your ADC, and the other from two decimations, one at the ADC and the
other fractional by software. If the first one, that was integer-decimated
in one pass, sounded worse can only mean one of two things (or both): the
analog antialiasing filter is not perfect for 44K1 sampling. The converter
uses the same analog filter for all sample rates. If this converter doesn't
attenuate enough the band beyond the original Nyquist freq (some MHz), then
sampling at higher rates can make a big difference as the original Nyquist
rises, but even more because the sonic power falls as the frequency is
higher. The other factor is the digital antialiasing in the ADC decimator.
If it doesn't attenuate aliases better than 120 dB, the difference can be
be noticed too.
Again, the difference can only be notified if the hardware quality doesn't
reach certain level (total aliasing rejection above final Nyquist in the
analog and digital filters better than 120 dB). And so, one more time, it's
not necessary to use higher sample rates to obtain the same quality, just
better converters. I don't mean that Sek's ones are bad, just not good
enough to make higher sample rates not noticeable.
I also want to comment other possible reasons (some has been suggested
already), but my oppinion is that they didn't influenced your experiment in
a meaningful measure:
3) Ultrasonic sound beyond 20 KHz can affect the audible range. It's true,
and has been demonstrated that several unaudible ultrasonic beams can create
audible sound when interfering in the air. This is bacause unlinearities in
the sound transmission properties of air. But I believe that this effect
would be negligible in your case because it's necessary high ultrasonic
power to influence a little, and surely your speakers and more things in the
chain doesn't perform well beyond 20 KHz. And in any case, this kind if
interferences would decrease quality, what is not what you found.
4) The difference in phase of signals reaching our ears is perceived as
positioning information. Higher sample rates improves phase accuracy. But
before reaching 44K1 the improvement is negligible compared with the phase
shifts in any speaker or analog crossover. The analog antialiasing filters
for ADCs also presents phase shifts near the cutoff corner, but with
oversampling converters, the designer has no problems to push up this
frequency to achieve linear phase up to 20 KHz.
As conclussion, based in all I know, there is not necessary to sample beyond
44K1 to achieve the same quality. I agree with you that storage and
thoughput is not a big constraint now, but twice sample rates, unlike higher
bit depths, requires twice CPU processing. And this is a resource that we
don't have in excess. At least for PC DAW, it's better to invest in better
converters than resorting to higher sample rates. In dedicated gear, it's
often more convenient using higher sample rates, because it's usually
cheaper in this case than using better converter.
Finally, if you, anybody, wants to measure or evaluate each one of your
converter properties, I'll be happy to explain how to do it easily. All you
need for the more important measurments is a signal generator (no quality
required), and an adequate sound editor (sure you have it <g>). I would also
explain rules to do comprehensive listening tests that isolates one factor
each time, and valid blind testing methods. That's to achieve meaningful
results.
Hope that helps,
--
J.M.Catena
ad...@sesa.es
George Brickner
sounding cards are probably equally resposable for the diferences
heard? Maybe more so.
On Wed, 23 Sep 1998 20:59:34 +0200, "Catena" <ad...@sesa.es> wrote:
>Dear Jim,
>Since I read in your past message that you found differences recording at
>96k, I wanted to know why, as I knew it isn't becasue what you think. Now,
>after reading your good article at www.prorec.com , I can say you why your
>conclussions are not valid.
>
>You compared a 96K recording converted to 44K1 with a direct 44K1 recording,
>didn't you? Well, what you in fact have "almost" demonstrated, is that 96K
>doesn't make any difference. Let me explain:
>
>1) When you convert 96K to 44k1, you don't notice any difference, right? It
>demonstrates that 96K doesn't improve sound.
>If you find any difference (very improbable), the reason is that the 96K to
>44K1 conversion introduces some distortion. Additionally, maybe the software
>isn't filtering above 22k05 well enough.
>To do a really accurate test, convert 96K to 48K, or 88K2 to 44K1 (to avoid
>distortion in the conversion) using software with a high performance
>antialiasing filter (to avoid aliasing). High performance here means, for
>example: 0.1 dB max ripple in the passband (20Hz-20KHz), >= 120dB stopband
>attenuation, constant phase (FIR filter with odd taps). That is, like the
>filters found in top quality 24 bit ADCs. In this case, you'll verify that
>96K sounds exactly the same as 48K. In fact, a 44K1 converter samples
>typically at 128x44K1, and then converts the rate digitally using a high
>performance antialiasing decimator, with characteristics as described.
>Anyway, you might agree that even with the weird 96K to 44k1 conversion you
>did, 96K doesn't improve sound.
>
>2) When comparing the 44K1 converted file with the directly recorded 44K1
>file, what are you comparing is not the difference between sample rates (as
>are already the same), but the antialiasing characteristics of your
>converters. From the results you got (difference appreciated), I can say
>that the antialiasing performance of these converters doesn't seem as good
>as you expected. If you want to measure objetively the antialiasing
>performance, let me know and I'll give you a test guide (you need only a
>signal generator capable of injecting freqs above Nyquist, quality isn't
>important for this). It's easy.
>
>Theory also serves to understand listening tests. Without theory, it's easy
>to deduct wrong conclussions. Theory and experiences are both necessary and
>complementary.
>
>I hope you understood everything. You can also find some background in my
>article at www.prorec.com (DSP theory). Don't hessitate to ask me if you
>have further
>doubts.
Chris Townsend
> Have any of you actually USED a set of 96k converters?
>
> Have YOU varified ANY of your hypothesis before claiming I'm full of
shit?
> I mean... with REAL working converters??? If so, which ones?
>
I've listened to the new DSD system on a very highend system, with source
material that should expose the limits of a CDs 20kHz bandwidth. So far I
wasn't able notice any real difference between the DSD recording and the
same material in CD format, which was produced from the original DSD
recording by performing high quality down sampling and dithering. This is
certainly not to say that there aren't some people who could tell the
difference though. And if I do much more in depth listening test, maybe I
could tell the difference in some cases.
> If not, here is your no-risk chance. <g>
> Pick up a set of the 2496s converters from me... and if you don't agree
with
> what I've said, I guarantee your money back. I can offer NO better proof
> than this.
I'd love to give'm a listen, but I'll probably wait until the price drops a
bit. Or maybe I'll ask Sek'D to give me a good demo.
>
> There is NOTHING I'm going to say that will change your mind...
> And by the same token, words typed here aren't going to change what I
hear
> when recording at 96k. And I do hear a difference... and so do others...
> Mo Weston runs a HUGE facility is Q8. He is actually USING 96k audio...
and
> he also hears a difference. Now, that's just a little odd, the guys who
USE
> the tool make observations, and the guys NOT using the tool try to
dispute
> what we hear.
>
I think you missed my point entirely, although correct me if I'm wrong. I
never argued that 96k sample rate is not better than 44.1k. 96K may be an
improvement, and I can think of a number of possible reasons why.
Nonetheless, I just argued that when doing listening tests it is very
difficult isolate each factor. So was the difference that you heard due to
the increase in sample rate or some other factor, like a change in the
amount of jitter of the converter? Just because the same converters and
system were used for the 44.1 and 96 sample rates, you won't necessarily
guarantee that all other things will be equal. And even if in this case
the Sek'D converter at 96k is truly an improvement over 44.1k due to, for
example, less aliasing artifacts etc. it doesn't mean that there aren't
other converters out there that can sample at 44.1kHz yet not suffer from
the same aliasing artifacts due to improved anti-aliasing filters. In my
opinion it is still too early to proclaim that 96k is in GENERAL an
improvement over 44.1k. Also, we may find that 96k is unnecessarily high,
and that 88.2k or 60k sample rates will be more than sufficient. Of course
others have suggested that 192k is necessary, but then why not 1MHz?
Also, you mentioned in another email that there are numerous people high up
in the industry touting 96k, to which I certainly agree, but there are also
many others who believe that 4416 or maybe 4420 is more than sufficient for
consumer audio rates. And of course unlike bit depth, there may not be
much advantage to using higher bit rates during recording and then down
sampling. Certainly the debate will rage for a long time.
Chris
--------------------------------
Chris Townsend - DSP Engineer
Arboretum Systems, Inc.
http://www.arboretum.com
--------------------------------
Catena
>Perhaps a converter 'shoot-out' would make a good prorec article.
It's coming <g>
--
J.M.Catena
ad...@sesa.es
John Harragin
seen so far I would not be afraid to pit this against sekd's little 96k
box.
John
"Catena" <ad...@sesa.es> writes: > >That was very helpful indeed!
John Harragin
In CW8 do we get better performance with 20 or 24 bit audio files? I
point this out because the 24 bit converters that us humans can afford
will have nothing but junk in the 4 least significant bits - with either
44.1 or 96k sample rates. I personally am willing to sacrifice a little
disk space for perfomance if that is the way this falls.
(Still I side with 100 years of accoustic reseach and am in the 44.1 camp
in the 44vs96 issue)
John
Jim Roseberry
>I have the oversampling 44.1/48k MultiWave pro analog 24. From what I've
>seen so far I would not be afraid to pit this against sekd's little 96k
>box.
I've used the Multiwave 24 analog and it does sound pretty good.
However... the 2496s produces results that sound better to my ears.
Understand... That isn't a cut on the Multiwave. I EXPECT that a $700
(street) set of converters would sound better.
Jim Roseberry
Studio Cat Software (Audio software/hardware sales)
Johnny Smooth
I just reread the tech article in EM. Not surprisingly I was
disappointed with it. :-) The problem is that the plot they
show of anti-aliasing filters responses 1x sampled data.
The filters they have are from 20K to ( 22.05, 24, and 48 )K.
True, the filter "ringing problem" improves with higher sampling
rates, so why did they leave out 5.6M, the 128x oversampled
rate? That gives a lot of room (20K to 2.8M) to avoid ringing.
They presented data that wasn't even applicable. That brick
wall looks more like a sloped driveway. One thing I haven't
seen specified anywhere is the oversampling rate for the
96K converters. If you could get that, I think it would be
helpful to the discussion. In the meantime, I'll try to find
some converters to listen to. :-)
Thanks,
John
pete leoni
>converters than resorting to higher sample rates. In dedicated gear, it's
>often more convenient using higher sample rates, because it's usually
>cheaper in this case than using better converter.
That's my opinion as well, I'm sure it is possible to get equally fine
results using both methods. It's just that the high sample rate, lower
quality filter route is so wasteful of resources, compared to the inverse.
We would all benefit greatly from low cost high quality converters, and if
we embrace the more expedient, yet less elegant high sample rate route, we
will lose in the long run. If one good thing has emerged from this exchange
it is the fact that the need for high quality, low cost converters at last
been noted and discussed.
pete
Catena
>seen specified anywhere is the oversampling rate for the
>96K converters.
Usually 64x or 128x. The CS5396 is 128x (24 bit, 100Ksps max out, DR=120 dB,
alias=-117dB, ripple=0.005 dB), but only $$$ units wear it for now.
--
J.M.Catena
ad...@sesa.es
John Vernon
>only $$$ units wear it for now.
During an exchange with Rob Ranck of Gadget Labs about the new Wave
8/24, I asked some questions about this: here is his response:
quote
Here's the facts...
http://www.cirrus.com/products/overviews/cs5360.html
unquote
So it may shortly not be all that many $$$...
John VERNON
Musique Animation France
JohnV...@compuserve.com
pete leoni
pete
John Vernon <JohnV...@compuserve.com> wrote in article
<361294f9...@news.cakewalk.com>...
Johnny Smooth
Catena wrote in message <6utt67$i...@hope.harvard.net>...
>>One thing I haven't
>>seen specified anywhere is the oversampling rate for the
>>96K converters.
>
>Usually 64x or 128x. The CS5396 is 128x (24 bit, 100Ksps max out, DR=120 dB,
>
Just looked at the data sheet, and if I read it correctly, 128x is only
available
for 48KHz; 64x is used for 96KHz. Which means both modes are really
sampling at the same rate.
John
Catena
worse. In fact, the 5360 performance is the theorical max for 18 bit
converters, not too much for a 24 bit one...
--
J.M.Catena
ad...@sesa.es
r...@gadgetlabs.com
"Catena" <ad...@sesa.es> wrote:
> The good ones are the 5394 and 5396. That 5360 is much cheaper... and much
> worse. In fact, the 5360 performance is the theorical max for 18 bit
> converters, not too much for a 24 bit one...
>
> --
> J.M.Catena
> ad...@sesa.es
Hi Catena,
Yes, it's certainly true that the Crystal 5394 and 5396 are better, more
expensive chips. But the 24-bit 5360 is better than the 20-bit chips found in
many PC audio cards.
Best regards,
Rob Ranck
Gadget Labs, Inc.
r...@gadgetlabs.com
-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
Catena
>available
>for 48KHz; 64x is used for 96KHz. Which means both modes are really
>sampling at the same rate.
Right. It's the reason of why I said that a 96K converter working a 48K runs
at half original sample rate than a 48K one, but this converter can work at
the max original sample rate with both modes thanks to the selectable
oversampling ratio, that is, it doesn't penalize 48K performance for the
capability to work at 96K.
The fact is what limits the max oversampling ratio is the maximum speed of
the sigma delta modulator, what is fixed for a given technology, being the
maximum original sample rate a constant.
--
J.M.Catena
ad...@sesa.es