It is currently Tue Mar 19, 2024 1:27 am


Selecting Bit Resolution at Organ Loading

A discussion forum for anything even marginally Hauptwerk-related.
  • Author
  • Message
Offline
User avatar

mdyde

Moderator

  • Posts: 15435
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Selecting Bit Resolution at Organ Loading

PostTue Mar 16, 2010 5:50 am

Hello Vidarf,

So perhaps some sample sets are more sensitive to 16 vs. 24 bit?


Yes - it depends on the signal level the producer chose for the samples - see my post earlier in this thread. MDA sets usually have a relatively low signal level so that those people with smaller/older computer systems have the option to load them in full within the smallest possible amount of memory (because lower signals levels compress more). But to get the best quality you need to load them in 24-bit (or 20-bit).

What you have stated above suggests that the cleanest organ sound should come from a dry set that has some sort of reverb added to it. Is that fair to say?


Hello Leo,

Potentially yes, in terms of noise levels (although dry sample sets are recorded close to the pipework which might make different sources of noise more prominent, e.g. air leakage or the blower).

That is, the dry pipes would not have much, if any, background noise that comes through at the end of the sample. There wouldn't be the build up of noise from many pipes playing together.


Correct.

A good reverb wouldn't add very much noise, would it?


Correct.

Perhaps 16-bit would do in this case as well as 24-bit?


Yes - for dry sample sets there's probably much less need for 24-bit. The signal levels in dry sample sets are often set higher than in wet ones anyway, since dry sample sets usually don't need as much memory as wet.
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline
User avatar

Disorganised

Member

  • Posts: 112
  • Joined: Tue Oct 09, 2007 6:13 pm
  • Location: London UK

Re: Selecting Bit Resolution at Organ Loading

PostTue Mar 16, 2010 7:00 pm

Well, having taken Jim's advice, I've just tried the Salisbury Cathedral sample set in 24,16 and 14 bit - and I can't hear any perceivable difference in sound quality at all.
John Mackay

London
UK
Offline
User avatar

Jim Reid

Member

  • Posts: 1764
  • Joined: Mon May 17, 2004 6:50 pm
  • Location: Kauai, Hawaii

More about .wav file sample bit loading

PostWed Mar 17, 2010 2:55 pm

Seems a good idea to break this info out into a separate topic from
the other thread which wandered off into other subjects.

Joe Hardy, also a member of this Forum, has posted at the VTPO
Yahoo list the following information and comments:

"From an email discussion with Ethan Wigner (posted with his permission):

""Even aside from any wind noise, no room you record in has a dynamic
range approaching 96 dB unless maybe you stuffed the microphone into
the pipe openings. [ed: in which case one would get some real noise!]

The main noise that will accumulate from playing many samples at once
is the background noise in each individual Wave file sample. Which, again,
I assume will be dominated by the room's ambient noise level.

A blind test is always useful, but this can be proven more easily simply by
reading the residual noise level on a VU meter for portions of the samples
where there's only silence. Are the people in this forum creating organ samples?
All you need is one 24-bit five-second clip of the room tone with no organ sounding.
Make a copy of that file reduced to 16 bits, then load up maybe 20 tracks of each
into a DAW program and press play. If all 20 24-bit files go to one output bus, and
all 20 16-bit files go to another, just read the playback level meters for each bus.
If the residual noise for both buses is the same, that proves the lower *potential*
noise from using 24 bits is not needed or being used.""

The 96 dB headroom provided by 16 bits is far more than the best professional
studio analog tape recorders which could muster perhaps 75 dB, not quite as
good as 13 bit resolution. Then you have the ambient noise which exists in
even the quietest rooms, the self-noise of the microphone, the noise in the
mic preamps, all of which gets included in the raw samples. Then of course
there is the organ wind noise.

It seems possible that a very quiet wet sampled classical organ may benefit from
loading in 24 bits when a large number of pipes are played and the room's natural
reverb decays fully. A test that would demonstrate whether or not there is any
benefit to the higher bit rate could be accomplished by loading the organ in 24 bits,
pull every available stop and coupler, open the swells, briefly play an eight note chord
and let the sound decay until the reverberation has completely decayed. Record this
directly to a WAV file using Hauptwerk's facility. Repeat with the organ loaded in 16 bits.
If there's a difference, it should be audible and measurable.

I do hope that everyone who has the slightest interest in this subject matter will watch
the entire 60 minute AES Audio Myth Workshop video:

http://www.youtube.com/watch?v=BYTlN6wjcvQ

The demonstrations of "expectation bias" are especially powerful."

End of quotes from Joe Hardy, and his from Ethan Wigner.

I have had the large 93 rank "wet" PAB sample set loaded in HW using
the "default" 16 bit loading with only the first sample loops loaded.
The result is a fine organ, but I would prefer to also have all the
sample loops. But, my 12 GB of RAM are too few.

So, this morning I am re-loading the PAB in only 14-bit and with all
the sample loops as well. With 16-bit loading, I am using 66% of the RAM
required to load at 24 bit (or I would have needed just over 18 GB
to load in 24 bit and still only one sample loop. With 14 bit loading, I
use 8% less RAM for the sample storage, or about 1 GB more.

I reasoned that this 1 GB more available RAM would allow me to again
try to load all the sample loops provided by Inspired Acoustics of the
organ pipes in this very large organ. It is now loaded; I will go play
some music now and report back about my satisfaction, or not with
this experiments results. If Ethan Wigner's arguments are correct,
the organ should sound just fine.

By the way, I believe the reason that his work and conclusions are
correct is that noise quantities do NOT add coherently, as one
learns when going into information transfer topics. At least,
as I believe I was taught so long ago in school.
Jim Reid
Offline
User avatar

Jim Reid

Member

  • Posts: 1764
  • Joined: Mon May 17, 2004 6:50 pm
  • Location: Kauai, Hawaii

Re: More about .wav file sample bit loading

PostWed Mar 17, 2010 3:38 pm

With the PAB samples loaded as described, 10.7 GB of RAM
is used.

Playing the organ, it sounds wonderful! No difference at all
from the 16 bit loading. And, pulling every stop ON of the
Great division and holding down two complete hands full (palms
pressing keys also) no noise could be heard even at the end
of the wet reverb tail as the complete fade point was approached,
where it should be most noticeable if it were to be, per Wigner.

I don't have the playback level meters needed to perform
the complete test as described previously. Perhaps someone
else does and could so perform and report.

Very interesting topic, at least for me.
Jim Reid
Offline
User avatar

mdyde

Moderator

  • Posts: 15435
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Selecting Bit Resolution at Organ Loading

PostWed Mar 17, 2010 3:55 pm

Hello Jim,

I've merged your new topic with the original one to keep the discussion in one place (it doesn't seem to fall within the category of Hauptwerk technical support, where you had it).
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline
User avatar

mdyde

Moderator

  • Posts: 15435
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Selecting Bit Resolution at Organ Loading

PostWed Mar 17, 2010 4:30 pm

Hello again Jim,

Why not just measure it for yourself, if you want to see whether there is actually any difference between loading a particular sample set in 14-bit vs. 24-bit?

E.g.:

1. Use a computer that's capable of a good polyphony (quad-core or better), the Hauptwerk Advanced Edition, and make sure the polyphony limit is set correctly for the computer in Hauptwerk (so that there's plenty of polyphony available to make it a meaningful test).

2. Make sure Hauptwerk is configured with just a single stereo output, and that the recording format is set to 32-bit on the 'General settings | Audio outputs' screen (so that you can detect and measure any differences in recorded noise levels as accurately as possible).

3. Choose one of the large and very wet sample sets that people have reported hearing a significant difference with in this thread (e.g. SP Caen or MDA Metz).

4. Load the sample set with all pipe ranks set to 24-bit. Also make sure you disable the blower 'rank', all action noise ranks and any other non-pipe ranks.

5. Draw the largest registration possible whilst avoiding the loud stops (since quieter stops are likely to show background noise more clearly). If that doesn't come close to the polyphony limit, use all stops instead. Draw all couplers. Write down the registration.

6. Use a piece of wood or similar to press all 61 keys on the main keyboard then carefully release them (release them all in exactly the same instant).

7. Wait 20 seconds (so that all reverb tails have had plenty of time to die away) then stop Hauptwerk recording.

Now repeat steps 4 to 7 but selecting 14-bit for all pipe ranks (and keeping all non-pipe ranks disabled) and using exactly the same registration that you used the first time.

Now exit Hauptwerk and open the two recordings in a good-quality audio editor, such as Sony Sound Forge, Adobe Audition or Steinberg WaveLab.

If loading in higher resolutions makes a measurable difference you would expect to see the mean amplitudes being higher in the 14-bit than in the 24-bit just before the release samples actually end. You might also expect to see a broader and more even spread of frequencies due to any additional noise. Work out the difference in mean amplitudes for segments of equal length a given time after the moment of key release, when the release samples have nearly but not quite finished playing using the audio editor's analysis functions (e.g. 'Tools | Statistics' in Sound Forge and/or FFT).

(There are plenty of better and more accurate ways you could analyse the files, but that should give a quick and easy rough estimate.)
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline
User avatar

Jim Reid

Member

  • Posts: 1764
  • Joined: Mon May 17, 2004 6:50 pm
  • Location: Kauai, Hawaii

Re: Selecting Bit Resolution at Organ Loading

PostWed Mar 17, 2010 6:38 pm

Martin, I could do the test as you suggest and make the two recordings.

However, I do not have Sony Sound Forge, Adobe Audition or Steinberg WaveLab.
I do have Sonar. Would that program perform the analysis you suggest? Or could
I send the two recordings to you to be examined? Even without modifying them
to .mp3 files, the .wav files would be of only 40 or 50 seconds duration performed
as you describe.
Jim Reid
Offline

jkinkennon

Member

  • Posts: 1208
  • Joined: Thu May 07, 2009 9:43 am
  • Location: Vancouver, WA

Re: Selecting Bit Resolution at Organ Loading

PostWed Mar 17, 2010 7:06 pm

My perverse sense of humor is to blame... Another test to support Martin's point that noise does in fact add up when you mix a thousand or so channels would be this. Try recording some "silent" tracks with microphones set up in the quietest room in your house. Perhaps eight tracks would be good. Then mix those down to two similar tracks and cut a bit from one channel, that is, offset them a bit just to be sure the noise stays non-coherent. Now mix these tracks and continue the process until you can hear the difference. Each doubling of samples is worth 3 db give or take, so by the time you hit 1024 samples in the mix it won't be quiet any more.

If you try this test at 24 and 16 bits you can't possibly not notice a difference. Now whether or not a good 24 bit recording sounds better than a 16 bit recording is quite another matter, but with Hauptwerk the key issue is where you start off before the mixing begins. Frankly, I'm in awe of how well the sampled organs work when I think of the noise issue.
Offline

jkinkennon

Member

  • Posts: 1208
  • Joined: Thu May 07, 2009 9:43 am
  • Location: Vancouver, WA

Re: Selecting Bit Resolution at Organ Loading

PostWed Mar 17, 2010 7:30 pm

Think of it this way. The sampling rate controls the high frequency response with 48k being good for a response up past 20 kHz. The sampling depth, typically from 16 to 24 bits and higher, controls the dynamic range. That's why it's probably correct to say that when all other factors are equal the 24 bit recording won't sound "better" in a blind test -- that is, unless it sounds quieter.

Actual room noise will predominate over thermal noise in any real world situation. But it still adds up in the same fashion. Sadly, the noisiest samples in the mix will have the largest influence on the noise present in the final mix.
Offline
User avatar

B. Milan

Site Admin

  • Posts: 4392
  • Joined: Sat Aug 30, 2003 10:15 am
  • Location: Sarasota, FL. USA

Re: Selecting Bit Resolution at Organ Loading

PostWed Mar 17, 2010 10:14 pm

Jim Reid wrote:Martin, I could do the test as you suggest and make the two recordings.

However, I do not have Sony Sound Forge, Adobe Audition or Steinberg WaveLab.
I do have Sonar. Would that program perform the analysis you suggest? Or could
I send the two recordings to you to be examined? Even without modifying them
to .mp3 files, the .wav files would be of only 40 or 50 seconds duration performed
as you describe.


Hello Jim,

In Sonar you can import an audio file and listen to it as a normal audio playback piece of software by going to File | Import | Audio after having first made sure to set up an audio track in Sonar and setting the sample rate properly in the Options | Audio menu (by default there should be at least 2 tracks when creating a new Sonar file). Make sure your audio card is chosen then just click the Play button to listen.
Brett Milan
Owner
MILAN DIGITAL AUDIO
Offline
User avatar

mdyde

Moderator

  • Posts: 15435
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Selecting Bit Resolution at Organ Loading

PostThu Mar 18, 2010 7:38 am

Hello Jim,

Sorry - I'm afraid I don't really have time to analyse them for you, but perhaps if somebody else here on the forum is interested they could do it for you if needed?

Another test to support Martin's point that noise does in fact add up when you mix a thousand or so channels would be this. Try recording some "silent" tracks with microphones set up in the quietest room in your house. Perhaps eight tracks would be good. Then mix those down to two similar tracks and cut a bit from one channel, that is, offset them a bit just to be sure the noise stays non-coherent. Now mix these tracks and continue the process until you can hear the difference. Each doubling of samples is worth 3 db give or take, so by the time you hit 1024 samples in the mix it won't be quiet any more.


If you have 4096 samples playing at once (=polyphony), which is quite often the case when using a very large, very wet cathedral organ sample set, then that's 12 'doublings' (4096 = 2^12)compared to a single sample, giving about 12 x 3 dB = 36 dB more noise/hiss due to quantisation noise effects than you would hear in a single sample.

With 16-bit samples the quantisation noise floor (maximum signal-to-noise ratio) is about -96 dB:

http://en.wikipedia.org/wiki/Quantization_error

20 x log10( 2^16) =approx= 96 dB

Hence if you play 4096 of them at once (as Hauptwerk would be doing when using a polyphony of 4096) the noise floor would rise to about -96 + 36 = -60 dB.

With 24-bit the noise floor is enormously lower:

20 x log10( 2^24 ) =approx= 144 dB

... so even with an additional 36 dB of noise resulting from a polyphony of 4096 the noise floor would only rise to about -144 + 36 = -108 dB.

The human ear has a dynamic range of about 120 dB:

http://www.dspguide.com/ch22/1.htm

The difference between the loudest and faintest sounds that humans can hear is about 120 dB, a range of one-million in amplitude. Listeners can detect a change in loudness when the signal is altered by about 1 dB (a 12% change in amplitude). In other words, there are only about 120 levels of loudness that can be perceived from the faintest whisper to the loudest thunder. The sensitivity of the ear is amazing; when listening to very weak sounds, the ear drum vibrates less than the diameter of a single molecule!


You probably won't generally be listening to your Hauptwerk organ at the loudest level that the human ear can distinguish (which I imagine would carry a risk of significant hearing damage!), but even at listening levels that re comfortably adjusted for maximum organ amplitude you should be able to hear a signal that is 60 dB quieter (i.e. at the approximate level of the quantisation noise floor you would get from playing 16-bit samples with a polyphony of 4096).

You can demonstrate this yourself with Sound Forge or similar:

- Create a new 2-second file. Use 'Tools | Synthesis | Simple' to generate a 0 dB 500 Hz sine wave.

- Turn your headphones / audio interface volume up so that you can hear it a loud but still comfortable listening level. It's important to use reasonable-quality headphones (not speakers) and a reasonably good quality audio interface.

- Now use 'Process | Volume' in Sound Forge to adjust the level of the sound by -59.99 dB (SF doesn't quite go to -60 dB!).

- Now play it again. You should easily still be able to hear the signal, even though it's about 60 dB quieter. (I just tried it myself to verify.)

To perform any accurate listening tests for details like noise you really need to be using just a single stereo output in Hauptwerk, with a reasonable-quality audio interface and reasonable-quality headphones. Speakers (even studio monitors) and listening room acoustics mask a lot of the final detail, so headphones are vital for hearing details accurately.

The main reason for including a blower sound effect in sample sets is to mask the cumulative noise that's inevitable when hundreds or thousands of samples play at once. For example, in the St. Anne's sample set the blower noise is about -36 dB (relative to the maximum signal level), so even with a polyphony of 4096 (not that St. Anne's actually has that many pipes!) it should still easily drown out -60 dB of quantisation noise (and other inevitable accumulated noise artefacts resulting from noise reduction in the original samples).
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline
User avatar

mdyde

Moderator

  • Posts: 15435
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Selecting Bit Resolution at Organ Loading

PostThu Mar 18, 2010 8:23 am

P.S. One important point I forgot to make:

If you played 4096 pipe samples together (starting at the same time), and if all of those pipe samples were played at the same amplitude and had fully been normalised so that each was stored at the loudest possible level within a 16-bit WAV file then of course the level heard would also be about 36 dB louder than playing just one sample.

Hence for comfortable listening you would have to turn the overall output level down by 36 dB, which would result in the quantisation noise being turned down by the same amount, and giving the same net signal-to-noise ratio from the 4096 samples playing as with just a single sample.

Joe Hardy is probably mainly talking about dry sample sets (since his interest in theatre organs), and for a dry sample set that could be fairly close to what happens in practice.

However, for a very wet sample set the situation is very different because you can easily be using a polyphony of 4096 (4096 samples playing at once) even though only a fairly small number of pipes are sounding (because the relesase samples for the reverb tails continuing playing from the last chord long after you've released that chord and played another one).

So say you have a maximum number of pipes actually sounding at once of:

10 fingers x 50 stops = 500 pipes =approx= 512

... but 4096 voices of polyphony being used (because of reverb tails from previous chords still playing).

512 = 2^9 whereas 4096 =2^12 so assuming that you adjusted the overall listening level to be comfortable with 512 pipes sounding, then the effective noise floor for 16-bit samples would rise compared to a single sample by about

3 dB x (12-9) = 9 dB.

If the pipe samples were all stored at maximum amplitude (not usually the case - see below) within their WAV files then that would give an effective noise floor rise from about -96 dB to about -96 + 9 = -87 dB.

If I try the same experiment as in my previous post (using Sound Forge to generate a 0 dB 500 Hz sine wave in a 32-bit audio file then turning it down by a given number of decibels, I find that the quietest (500 Hz) signal I can reliably hear (relative to a loud sound at comfortable listening level) is about -84 dB.

I.e. my hearing (and/or the headphones and/or audio interface I'm using at the moment) isn't quite good enough to hear a -87 dB signal (unless I turn it up again of course!), so in the above example the quantisation noise floor would be acceptable (if hearing was equally sensitive to all frequencies).

However, if you had the overall output level adjusted for playing less than 512 maximum-amplitude samples then of course the noise floor would be higher relatively. In the case of my own hearing (if all samples were normalised for maximum amplitude) then I would need to have the level set comfortably for about 256 or less pipes actually sounding at once (as distinct from polyphony) before the 16-bit quantisation noise should theoretically become audible to me.

The other very important factor is the signal level within the samples. Most wet sample sets have the relative signal level between samples (or at least across the compass within each given rank) kept approximately as it was recorded for several reasons:

1. It automatically preserves the original relative amplitudes of the pipes (without needing to normalise the amplitudes of each one and then turn them down again in the organ definition). So it helps to reduce sample set production time/cost.

2. A lower signal level means that the sample set can be loaded in less memory in 16-bit and 14-bit (since lower signal levels compress significantly better in memory), so the overall hardware requirements can be lower for those with smaller/older computers (those with more memory can simply load in 20-bit or 24-bit to recover the full signal-to-noise ratio) than they would be with maximum signal-level samples.

3. It makes the samples better compatible with the Custom Organ Design Module (CODM), and compatible with the CODM in the Basic Edition (otherwise the user would have to use the Advanced Edition's voicing facilities to adjust the amplitude of every pipe back to a guessed appropriate level).
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline
User avatar

Jim Reid

Member

  • Posts: 1764
  • Joined: Mon May 17, 2004 6:50 pm
  • Location: Kauai, Hawaii

Re: Selecting Bit Resolution at Organ Loading

PostThu Mar 18, 2010 1:02 pm

Martin.

Thank you for your very detailed replies!

I understand them also. I have found that Sound Forge Studio
can be purchased for about $55 now at Amazon. I just might do so, but not
immediately -- have just paid Fed and county taxes!

Have found that my hearing (age 75 now) is essentially gone above
about 8 kHz, Perhaps that aids in my lack of noise sensitivity at very
weak levels. But my hearing sensitivity is still excellent
up to some 7 kHz; I do hear very faint sounds outdoors, leaves
rustling, frogs croaking from a distant reservoir, at least at
night.

I use Austrian made Reference headphones, K701 and have four
test CDs, three by "stereophile" and one with frequency test tracks
from 32 to 20,000 Hz, Rives Test CD 2 .

Thanks again for your complete and comprehensive help.
Jim Reid
Offline
User avatar

mdyde

Moderator

  • Posts: 15435
  • Joined: Fri Mar 14, 2003 1:19 pm
  • Location: UK

Re: Selecting Bit Resolution at Organ Loading

PostThu Mar 18, 2010 2:32 pm

Thanks, Jim.

No problem.
Best regards, Martin.
Hauptwerk software designer/developer, Milan Digital Audio.
Offline

Eric Sagmuller

Member

  • Posts: 911
  • Joined: Fri Apr 11, 2008 4:18 pm
  • Location: Bellefonte, PA USA

Re: Selecting Bit Resolution at Organ Loading

PostThu Mar 18, 2010 5:27 pm

I've been running Bovenkerk on 24 bit, just volume 1 so far, I have the rest just haven't taken the time to install them yet.

Anyway, when the reverb fades away there is like this swishing sound towards the end before it becomes silent. Even with the blower I can hear it, but especially without. I'm curious if this would be worse with 16 bit.

On the other hand the Freiberg demo I have from Jiri fades out silently without any of this noise. He claims he has a special de-noising technic?

Eric
PreviousNext

Return to General discussion

Who is online

Users browsing this forum: No registered users and 5 guests