The sample rate/bit depth war

[Carey Carlan]

I really don't want to start this battle again, but I'm trying to justify
to my video counterparts that 48K/16 sampling is adequate for a blu-ray
disk that can handle 96K/24.

Here's the test I'm planning:

1) Let them choose a blu-ray soundtrack that they think exemplifies the
niceties of 96K sampling.

2) Rip that sample bit-accurate into 96/24.

3) Create an disk from the ripped copy still in 96/24.

4) Reduce the level by 48 dB then raise it back again, effective emptying
the bottom 8 bits.

5) Create another disk as in (3) above at 96/24.

6) Have them pick between disks from (3) and (5) in a blind test.

By using copies made from the same source using the same steps and playing
back in the same machine with the same converters, I should get a fair
test. There is a pass of decompression/recompression when reading/creating
a blu-ray disk, but both disks will have the same processing.

Anything wrong with this scenario?

[Les Cargill[_2_]]

Carey Carlan wrote:

> I really don't want to start this battle again, but I'm trying to justify
> to my video counterparts that 48K/16 sampling is adequate for a blu-ray
> disk that can handle 96K/24.
>
> Here's the test I'm planning:
>
> 1) Let them choose a blu-ray soundtrack that they think exemplifies the
> niceties of 96K sampling.
>
> 2) Rip that sample bit-accurate into 96/24.
>
> 3) Create an disk from the ripped copy still in 96/24.
>
> 4) Reduce the level by 48 dB then raise it back again, effective emptying
> the bottom 8 bits.
>


Maybe, maybe not - I'd look at the samples first.

> 5) Create another disk as in (3) above at 96/24.
>
> 6) Have them pick between disks from (3) and (5) in a blind test.
>
> By using copies made from the same source using the same steps and playing
> back in the same machine with the same converters, I should get a fair
> test. There is a pass of decompression/recompression when reading/creating
> a blu-ray disk, but both disks will have the same processing.
>
> Anything wrong with this scenario?


Take both disks, rip 'em, calculate the difference signal* and
see.

*in CoolEdit, this involves copying one disk's worth, and using
the "mix paste" tool with the 'invert' checkbox selected.

--
Les Cargill

[RD Jones]

On Nov 14, 4:22*pm, Carey Carlan > wrote:
> I really don't want to start this battle again, but I'm trying to justify
> to my video counterparts that 48K/16 sampling is adequate for a blu-ray
> disk that can handle 96K/24.
>
> Here's the test I'm planning:
>
> 1) Let them choose a blu-ray soundtrack that they think exemplifies the
> niceties of 96K sampling.
>
> 2) Rip that sample bit-accurate into 96/24.
>
> 3) Create an disk from the ripped copy still in 96/24.
>
> 4) Reduce the level by 48 dB then raise it back again, effective emptying
> the bottom 8 bits.
>
> 5) Create another disk as in (3) above at 96/24.
>
> 6) Have them pick between disks from (3) and (5) in a blind test.
>
> By using copies made from the same source using the same steps and playing
> back in the same machine with the same converters, I should get a fair
> test. *There is a pass of decompression/recompression when reading/creating
> a blu-ray disk, but both disks will have the same processing.
>
> Anything wrong with this scenario?

This only addresses the word length.

Some disks may have 48k sample rate disguised as 96k.
Examine the data to determine both the true sample rate and
word length of the audio.

Then you can do 3 tests, one for 96->48->96,
one for 24->16->24, and the big one ...
96/24->48/16->96->24.

Personally, I think 48/24 is more than adequate, and that
outside of a "monitoring" environment very few listeners will
discern anything better than 48/16. JMO.

I would have liked CDs to adopt 48/16 at the beginning,
but it's way too late to talk about that now.

rd

[sTeeVee]

On Nov 14, 9:26*pm, RD Jones > wrote:
> On Nov 14, 4:22*pm, Carey Carlan > wrote:
>
>
>
>
>
> > I really don't want to start this battle again, but I'm trying to justify
> > to my video counterparts that 48K/16 sampling is adequate for a blu-ray
> > disk that can handle 96K/24.
>
> > Here's the test I'm planning:
>
> > 1) Let them choose a blu-ray soundtrack that they think exemplifies the
> > niceties of 96K sampling.
>
> > 2) Rip that sample bit-accurate into 96/24.
>
> > 3) Create an disk from the ripped copy still in 96/24.
>
> > 4) Reduce the level by 48 dB then raise it back again, effective emptying
> > the bottom 8 bits.
>
> > 5) Create another disk as in (3) above at 96/24.
>
> > 6) Have them pick between disks from (3) and (5) in a blind test.
>
> > By using copies made from the same source using the same steps and playing
> > back in the same machine with the same converters, I should get a fair
> > test. *There is a pass of decompression/recompression when reading/creating
> > a blu-ray disk, but both disks will have the same processing.
>
> > Anything wrong with this scenario?
>
> This only addresses the word length.
>
> Some disks may have 48k sample rate disguised as 96k.
> Examine the data to determine both the true sample rate and
> word length of the audio.
>
> Then you can do 3 tests, one for 96->48->96,
> one for 24->16->24, and the big one ...
> 96/24->48/16->96->24.
>
> Personally, I think 48/24 is more than adequate, and that
> outside of a "monitoring" environment very few listeners will
> discern anything better than 48/16. JMO.
>
> I would have liked CDs to adopt 48/16 at the beginning,
> but it's way too late to talk about that now.
>
> rd- Hide quoted text -
>
> - Show quoted text -

Take this drivel and shove it - I'm not readin this no more.

[Arny Krueger]

"Carey Carlan" > wrote in message


> I really don't want to start this battle again, but I'm
> trying to justify to my video counterparts that 48K/16
> sampling is adequate for a blu-ray disk that can handle
> 96K/24.

> Here's the test I'm planning:

> 1) Let them choose a blu-ray soundtrack that they think
> exemplifies the niceties of 96K sampling.

> 2) Rip that sample bit-accurate into 96/24.

By what means that avoids the copy protection?

> 3) Create an disk from the ripped copy still in 96/24.

> 4) Reduce the level by 48 dB then raise it back again,
> effective emptying the bottom 8 bits.

Again, by what means? Emptying the bottom bits with common audio software
is not always trivial because they do their work in 32 bit floating point.

My preferred method for emptying the lower bits is actually pretty obvious -
convert it to 16 bits and then convert it back to 24 bits.

This is basically the methodology described in several recent AES papers
including one by John Vanderkooy at the recent San Francisco meeting.

> 5) Create another disk as in (3) above at 96/24.

> 6) Have them pick between disks from (3) and (5) in a
> blind test.

> By using copies made from the same source using the same
> steps and playing back in the same machine with the same
> converters, I should get a fair test. There is a pass of
> decompression/recompression when reading/creating a
> blu-ray disk, but both disks will have the same
> processing.

> Anything wrong with this scenario?

This has been done many times and the results are that nobody hears a
difference unless they do something like listen to very faint passages with
high gain settings that would results in excess blasting during the loud
passages.

[Carey Carlan]

"Arny Krueger" > wrote in
:

> "Carey Carlan" > wrote in message
>
>
>> I really don't want to start this battle again, but I'm
>> trying to justify to my video counterparts that 48K/16
>> sampling is adequate for a blu-ray disk that can handle
>> 96K/24.
>
>> Here's the test I'm planning:
>
>> 1) Let them choose a blu-ray soundtrack that they think
>> exemplifies the niceties of 96K sampling.
>
>> 2) Rip that sample bit-accurate into 96/24.
>
> By what means that avoids the copy protection?

The program is called "AnyDVDHD".

>> 3) Create an disk from the ripped copy still in 96/24.
>
>> 4) Reduce the level by 48 dB then raise it back again,
>> effective emptying the bottom 8 bits.
>
> Again, by what means? Emptying the bottom bits with common audio
> software is not always trivial because they do their work in 32 bit
> floating point.

Hmm. Correct. In 32 bit, that would be a 96 dB cut and restore.

> My preferred method for emptying the lower bits is actually pretty
> obvious - convert it to 16 bits and then convert it back to 24 bits.

I was trying to avoid any data conversions. By doing nothing but
changing levels I might convince others that no magic was happening.

> This is basically the methodology described in several recent AES
> papers including one by John Vanderkooy at the recent San Francisco
> meeting.
>
>> 5) Create another disk as in (3) above at 96/24.
>
>> 6) Have them pick between disks from (3) and (5) in a
>> blind test.
>
>> By using copies made from the same source using the same
>> steps and playing back in the same machine with the same
>> converters, I should get a fair test. There is a pass of
>> decompression/recompression when reading/creating a
>> blu-ray disk, but both disks will have the same
>> processing.
>
>> Anything wrong with this scenario?
>
> This has been done many times and the results are that nobody hears a
> difference unless they do something like listen to very faint passages
> with high gain settings that would results in excess blasting during
> the loud passages.

I completely agree with the conclusions. I'm trying to convince those I
work with that my formats are adequate so I won't having to work in
extreme HD which is slower with bigger files and no benefit.

[Scott Dorsey]

Carey Carlan > wrote:
>I really don't want to start this battle again, but I'm trying to justify
>to my video counterparts that 48K/16 sampling is adequate for a blu-ray
>disk that can handle 96K/24.
>
>Here's the test I'm planning:
>
>1) Let them choose a blu-ray soundtrack that they think exemplifies the
>niceties of 96K sampling.
>
>2) Rip that sample bit-accurate into 96/24.
>
>3) Create an disk from the ripped copy still in 96/24.
>
>4) Reduce the level by 48 dB then raise it back again, effective emptying
>the bottom 8 bits.
>
>5) Create another disk as in (3) above at 96/24.
>
>6) Have them pick between disks from (3) and (5) in a blind test.
>
>By using copies made from the same source using the same steps and playing
>back in the same machine with the same converters, I should get a fair
>test. There is a pass of decompression/recompression when reading/creating
>a blu-ray disk, but both disks will have the same processing.
>
>Anything wrong with this scenario?

Make a third disk that is a 48/16 disk but with the levels bumped up by 1 dB.
Throw it into the mix.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Ethan Winer[_3_]]

On Nov 15, 9:32 am, Carey Carlan > wrote:
> I was trying to avoid any data conversions. By doing nothing but
> changing levels I might convince others that no magic was happening.

I do this all the time by loading a "high-res" file into Sound Forge,
then reducing the bit depth and saving that as a copy. This is more
direct than lowering and raising the gain, which may not do what you
think it does. Other than that, I agree with your general concept of
testing people blind.

--Ethan

[Mark]

>
> > This has been done many times and the results are that nobody hears a
> > difference unless they do something like listen to very faint passages
> > with high gain settings that would results in excess blasting during
> > the loud passages.
>
> I completely agree with the conclusions. *I'm trying to convince those I
> work with that my formats are adequate so I won't having to work in
> extreme HD which is slower with bigger files and no benefit.-
>
>
you never will convince them

these kind of audiophool things take the form of religious beliefs...

no amount of sceintific evidence will change their belief..

Mark

[Carey Carlan]

Mark > wrote in news:1fbb9ae4-f5b4-4492-9843-
:

> you never will convince them
> these kind of audiophool things take the form of religious beliefs...
> no amount of sceintific evidence will change their belief..
>
> Mark

The people I'm working with aren't prejudiced. They just don't know any
better.

[Carey Carlan]

(Scott Dorsey) wrote in
:

> Make a third disk that is a 48/16 disk but with the levels bumped up
> by 1 dB. Throw it into the mix.
> --scott

Oh, that's sneaky! I'll do just that.

[Arny Krueger]

"Carey Carlan" > wrote in message

> (Scott Dorsey) wrote in
> :
>
>> Make a third disk that is a 48/16 disk but with the
>> levels bumped up by 1 dB. Throw it into the mix.
>> --scott
>
> Oh, that's sneaky! I'll do just that.

Not sneaky, but an excellent double check. Some listeners just don't get it,
and failing to hear a 1 dB (or maybe 2 dB) overall level shift will expose
them.

If *everybody* fails to detect the 1 dB shift, there could be a problem with
the program material that you are using. I've seen normally good listeners
falter with program material that jumps around too much or is really
unfamiliar.

[Carey Carlan]

"Arny Krueger" > wrote in
:

> "Carey Carlan" > wrote in message
>
>> (Scott Dorsey) wrote in
>> :
>>
>>> Make a third disk that is a 48/16 disk but with the
>>> levels bumped up by 1 dB. Throw it into the mix.
>>> --scott
>>
>> Oh, that's sneaky! I'll do just that.
>
> Not sneaky, but an excellent double check. Some listeners just don't
> get it, and failing to hear a 1 dB (or maybe 2 dB) overall level shift
> will expose them.
>
> If *everybody* fails to detect the 1 dB shift, there could be a
> problem with the program material that you are using. I've seen
> normally good listeners falter with program material that jumps around
> too much or is really unfamiliar.

Sneaky in the sense that I've never heard an audio test that didn't prefer
the louder signal (within reason). I'm dealing with video people. I don't
believe they'd realize that louder has an advantage.

[Arny Krueger]

"Carey Carlan" > wrote in message

> "Arny Krueger" > wrote in
> :
>
>> "Carey Carlan" > wrote in message
>>
>>> (Scott Dorsey) wrote in
>>> :
>>>
>>>> Make a third disk that is a 48/16 disk but with the
>>>> levels bumped up by 1 dB. Throw it into the mix.
>>>> --scott
>>>
>>> Oh, that's sneaky! I'll do just that.
>>
>> Not sneaky, but an excellent double check. Some
>> listeners just don't get it, and failing to hear a 1 dB
>> (or maybe 2 dB) overall level shift will expose them.
>>
>> If *everybody* fails to detect the 1 dB shift, there
>> could be a problem with the program material that you
>> are using. I've seen normally good listeners falter
>> with program material that jumps around too much or is
>> really unfamiliar.
>
> Sneaky in the sense that I've never heard an audio test
> that didn't prefer the louder signal (within reason).
> I'm dealing with video people. I don't believe they'd
> realize that louder has an advantage.


Hmmm, while I am a video person of sorts, I was first an audio person...

The point is there though - some music makes technical differences far
easier to hear than others. For example, while there is little controversy
that in general an 11 KHz brick wall filter should be audible, it is
possible to find natural program material that makes it hard to detect.

OTOH, there are a very few musical instruments that basically sit there and
ring strongly around 18 KHz, no matter what notes you play. Just touch them!
Most people don't consciously hear the high pitched tone as such, but take
it away with a 16 KHz brick wall filter and many listeneners will notice its
comings and goings. Does the existance of rare instruments like these
justify the need to always carry the bandpass out to 20-22 KHz, or is it
reasonable to follow the general rule that holds for almost all music, which
is that clean brick wall filtering around 16 KHz is just fine? These are
the sorts of things that perceptual coder designers worry about.

My philosophy is that I pick reasonable worst case test samples that are
reasonably acessible to the people in the US. So, jangling key chains are
OK, and a zither would be OK, but odd traditional instruments used rarely
in a far corner of a lost province of Spain are not.

[Mr Soul]

On Nov 15, 7:21*am, "Arny Krueger" > wrote:
> "Carey Carlan" > wrote in message
>
>
>
> > I really don't want to start this battle again, but I'm
> > trying to justify to my video counterparts that 48K/16
> > sampling is adequate for a blu-ray disk that can handle
> > 96K/24.
> > Here's the test I'm planning:
> > 1) Let them choose a blu-ray soundtrack that they think
> > exemplifies the niceties of 96K sampling.
> > 2) Rip that sample bit-accurate into 96/24.
>
> By what means that avoids the copy protection?
>
> > 3) Create an disk from the ripped copy still in 96/24.
> > 4) Reduce the level by 48 dB then raise it back again,
> > effective emptying the bottom 8 bits.
>
> Again, by what means? *Emptying the bottom bits with common audio software
> is not always trivial because they do their work in 32 bit floating point..
>
> My preferred method for emptying the lower bits is actually pretty obvious -
> convert it to 16 bits and then convert it back to 24 bits.
>
> This is basically the methodology described in several recent AES papers
> including one by John Vanderkooy at the recent San Francisco meeting.
>
> > 5) Create another disk as in (3) above at 96/24.
> > 6) Have them pick between disks from (3) and (5) in a
> > blind test.
> > By using copies made from the same source using the same
> > steps and playing back in the same machine with the same
> > converters, I should get a fair test. *There is a pass of
> > decompression/recompression when reading/creating a
> > blu-ray disk, but both disks will have the same
> > processing.
> > Anything wrong with this scenario?
>
> This has been done many times and the results are that nobody hears a
> difference unless they do something like listen to very faint passages with
> high gain settings that would results in excess blasting during the loud
> passages.
Not sure what you mean by not hearing the difference? In my home
work, I can hear the difference between recording 44.1/24 and 96/24,
particularly with acoustic tracks, on my Sony headphones, with the
96/24 sounding better. Maybe this is because my audio card is of
medium quality but I can definitely hear the difference.

IMO - if you have a BR, then why not use the higher quality audio?
Having said that, I think I might question going to 192.

Mike C

[Arny Krueger]

"Mr Soul" > wrote in message


> Not sure what you mean by not hearing the difference?

I mean in a level-matched, time-synched, double blind listening test.

> In my home work, I can hear the difference between recording
> 44.1/24 and 96/24, particularly with acoustic tracks, on
> my Sony headphones, with the 96/24 sounding better.
> Maybe this is because my audio card is of medium quality
> but I can definitely hear the difference.

Like many others, I've tried this sort of thing with some of the finest
audio cards, loudspeakers, and headphones around. I've had a goodly number
of young, enthusiastic, well-trained listeners try it as well.

Post again should you have the resources to try a level-matched,
time-synched, double blind listening test.

Don't feel like you've been singled out. Everybody sings a different tune
after doing a proper blind listening test.

[Scott Dorsey]

Arny Krueger > wrote:
>Don't feel like you've been singled out. Everybody sings a different tune
>after doing a proper blind listening test.

Not always, in part because there are some lousy converters out there. I have
heard one case where the 44.1 audio sounded noticeably better than the same
system at 96.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Mr Soul]

> I mean in a level-matched, time-synched, double blind listening test.
OK - I see what you mean. In a real scientific test, I have no idea
how I would do. However, it my home studio, with material I was
familar with, I thought 96/24 sounded better.

Why the heck would anyone want to record in 192? Just to waste disk
space. That does seem like overkill to me.

Mr Soul

[Boris Lau]

On 11/16/2010 09:29 PM, Mr Soul wrote:
>> I mean in a level-matched, time-synched, double blind listening test.
> OK - I see what you mean. In a real scientific test, I have no idea
> how I would do. However, it my home studio, with material I was
> familar with, I thought 96/24 sounded better.
The way of testing makes a big difference. You can hardly the placebo
effect in such subjective tests.

The good news is that you can easily do such tests for yourself. Just
get a free abx comparison software like Foobar2000, abx comparator
(linux), winabx or pc-abx and check your scores...

Cheers,
Boris

[Anahata]

On Tue, 23 Nov 2010 01:24:45 -0800, Marc Wielage wrote:

>
> If the intended audience is audiophiles who care, then maybe 96/24 a
> selling point and worth the extra trouble.

Good point about this being a marketing issue, not entirely a technical
one. Whether or not anyone can hear the difference, there will be some
who want to buy 24/96 "because it's better"

--
Anahata
--/-- http://www.treewind.co.uk
+44 (0)1638 720444

[alex]

Il 14/11/2010 23.22, Carey Carlan ha scritto:
> I really don't want to start this battle again, but I'm trying to justify
> to my video counterparts that 48K/16 sampling is adequate for a blu-ray
> disk that can handle 96K/24.
>
> Here's the test I'm planning:
>
> 1) Let them choose a blu-ray soundtrack that they think exemplifies the
> niceties of 96K sampling.
>
> 2) Rip that sample bit-accurate into 96/24.
>
> 3) Create an disk from the ripped copy still in 96/24.
>
> 4) Reduce the level by 48 dB then raise it back again, effective emptying
> the bottom 8 bits.
No, you just need to empty the upper 8 bits, the less significative.
The best approach is still to TRUNCATE.
Otherwise your daw will increase the bithdepth to 32 (at least) and is
very hard to get the upper (16) equal to 0 by level adjustments.
By the way 48 is not good for this, just 7 bits will be free. the magic
number is 48,164799306236991234198223155919 dB, or 49! ;-)

>
> 5) Create another disk as in (3) above at 96/24.
>
> 6) Have them pick between disks from (3) and (5) in a blind test.
>
> By using copies made from the same source using the same steps and playing
> back in the same machine with the same converters, I should get a fair
> test. There is a pass of decompression/recompression when reading/creating
> a blu-ray disk, but both disks will have the same processing.
>
> Anything wrong with this scenario?
Nothing wrong in this comparision, but DVD and BD are discs. On discs
you loose what you don't use, so if the video part don't need more
space, feel free to use 96kHz 24bits.

Is much more useful to reduce the SR from 96 to 48kHz than the bitdepth!
with SR you'll reduce the size by 2 against the bitdepth where you
reduce just by 1.5.

For the SR part, the blind test is very hard to be made because people
don't know where to find differences, but there are still small
differencies!
96kHz will give you a better waveform shape on highs. This not always
noticeable.
96 make sense if you have a 96kHz recorded sound from the beginning.
Upconvert from 48 to 96 is meaningless.
The available room on disc and the max allowed program bitrate is the
best discriminator in order to take a decision.

The bitdepth is much more critical. 24 bits make a very big difference
against 16. Always try to use 24 (just a 50% increase in size, but 256
times the dynamic range, 48dB!). Even if your original material is 16,
internally, during the processing, the bitdepth is increased to 32 or
more (daws) and is better to dither down to 24 at the end of the process.

[Ben Bradley[_2_]]

On Mon, 15 Nov 2010 14:32:51 GMT, Carey Carlan >
wrote:

>"Arny Krueger" > wrote in
:
>
>> "Carey Carlan" > wrote in message
>>
>>
>>> I really don't want to start this battle again, but I'm
>>> trying to justify to my video counterparts that 48K/16
>>> sampling is adequate for a blu-ray disk that can handle
>>> 96K/24.
>>
>>> Here's the test I'm planning:
>>
>>> 1) Let them choose a blu-ray soundtrack that they think
>>> exemplifies the niceties of 96K sampling.
>>
>>> 2) Rip that sample bit-accurate into 96/24.
>>
>> By what means that avoids the copy protection?
>
>The program is called "AnyDVDHD".
>
>>> 3) Create an disk from the ripped copy still in 96/24.
>>
>>> 4) Reduce the level by 48 dB then raise it back again,
>>> effective emptying the bottom 8 bits.
>>
>> Again, by what means? Emptying the bottom bits with common audio
>> software is not always trivial because they do their work in 32 bit
>> floating point.
>
>Hmm. Correct. In 32 bit, that would be a 96 dB cut and restore.

That would be 32 bit fixed-point. In floating point a 96dB cut
might just change the exponent (essentially the binary point) and
still keep all the significant bits. Not sure of the exponent range,
but it's a bother figuring out what floating point range would clip
data by how much.

>
>> My preferred method for emptying the lower bits is actually pretty
>> obvious - convert it to 16 bits and then convert it back to 24 bits.
>
>I was trying to avoid any data conversions. By doing nothing but
>changing levels I might convince others that no magic was happening.

Reduce the level by 48dB, write it to a 24-bit external file, load
that file, then amplify by 48dB. This bypasses any internal
floating-point "magic" that prevents bit depth loss with gain change.