ORCHESTRAL POLYPHONY :
Updated 03/18/2003
A lot of people ask: how many samplers (EXS24, Halion, Kontakt) do I have to run to simulate an Orchestra?
With the new Sound Library formats (VLS, Sonic Implant,etc.) and samplers capable of handling bigger sample libraries, this question has become all the more relevant. Let's try to answer it.
First of all, let's try to answer the following questions : what is a regular orchestral structure and how many voices have to play simultaneously to reproduce such an orchestra?

A regular Orchestral Structure
We'll look at the different sections a regular orchestra has, and count the number of voices a sampler would need to reproduce each section.

a)	Small Harmony
Look at the first picture, showing a Typical Orchestral Structure. We see 2 flutes, 2 oboes, 2 clarinets, 2 bassoons.
Let's call this the small harmony structure.
Each instrument is monophonic, i.e. only 1 note at a time can be played.

How many notes can the small harmony have?
2 flutes: 	                  2 notes
2 oboes:	                  2 notes
2 clarinets:	                  2notes
2 bassoons:	2 notes
For one note played (this is important, you will understand why later ...)
What do you mean with the above sentence?  That the above table applies when all instruments play a single note?  But they always do, right, since they're monophonic?
So this is 8 notes maximum in Mono, or 16 notes in stereo. However, you'll need 32 notes if there is some "Release" on the ADSR, otherwise the next note will cut off the previous one.
Even though e.g. a flute can only play one note at a time, having the first sample (one note) get cut-off by the next sample (second note) tends to sound rather unrealistic. Things often sound a lot smoother and more natural when you allow notes to overlap a bit, or when you apply a little bit of 'release' to the notes. That means however that you then need double the number of voices. In mono you would thus use 16 note polyphony instead of 8, and in stereo this would be 32 notes instead of 16.
Let us, for the moment, assume we need 32 notes of polyphony. Note that using different articulations for one instrument doesn't increase the polyphony (fortunately): when playing e.g. staccato instead of legato, you simply replace one set of samples by another, which thus doesn't affect polyphony.
So: small harmony needs 32 notes 

b)	Horns
4 fhorns:	4 voices
2 trumpets:	4 voices since we let each trumpet part be played by 2 trumpet-players
3 trombones:	3 voices
1 bass trombone:	1 voice
2 tubas:	                  2 voices

The exact orchestral sound depends a lot on the number of musicians. For example, you can have 3 bass trombones or 3 tenors, etc. Ditto for the trumpets: even though there are just 2 trumpet lines, they can be played by 2 musicians each, thus using up 4 notes for the trumpets. The above overview is a sort of 'average', to which we'll stick for now.
So, we have 14 musicians/ 14 voices simultaneously (mono), 28 voices in stereo, and 56 voices (stereo) due to ADSR or slightly overlapping notes. That's a lot!
But you might very well have a sampler patch that is e.g. a complete "4 trumpets" program. In that case you don't need to use 4 voices for the trumpets, but only one. That way you can potentially cut back on polyphony quite a bit. The number of voices would then be as follows:
(for one note played, 4 musicians heard for the trumpets, no divisi ...)
Horns :		1 voice in mono, 2 stro, 4 with ADSR
trumpets:	1 voice in Mono , 2 voices in stereo , 4 voices with some ADSR
trombones:	1 voice in Mono, 2 voices  stereo, 4 with ADSR
bass trombone:	1 voice mono, 2 voices  stereo, 4 voices with ADSR
tuba:	1 voice mono, 2 voices stereo, 4 voices with ADSR
We now only need 20 voices instead of 56, in stereo with ADSR. That's much better!

In case of "divisi" ( different notes played by the same group of instrument) calculation would be:
4 (differents) horn lines:	            16 voices
2 (differents) trumpet lines:	8 voices
2 (differents) trombones lines:	8 voices
1 trombone bass line :                 4 voices
1 tuba:	                4 voices
40 notes voices.

So now we have:
Small Harmony:	32 notes
Horns:	20 notes
We now thus have a 52-voices structure.

c)	Keyboard and Harp
Keyboard and harp are very different from e.g. flute or a horn. First of all they are polyphonic (i.e. more than one note at a time can be played). Second of all, they have a damper pedal, which allows for huge release times, which of course also increases polyphony.
For a normal keyboard you'll need 32 notes in stereo mode, which for me it the minimun (but I'm a pianist) and 64 with some ADSR. The same holds for the harp (also because it is so beautiful with release).
So we need 128 Notes in stereo mode with some ADSR-release (not too much ...)

Small Harmony (32) + Horns (20) + Keyboard & Harp (128) = 180 Voices

What hasn't been covered yet?
- percussion
- strings section: violin 1&2 , altviola, celli, double bass
For now we'll skip the electric bass, guitar, etc. Let's try to get a simple orchestral structure.

d)	Percussion
Timpani (long release) = 8 notes due to roll possibilities
Snare ( short release) = 2 notes ( left/right hand)
Triangle (long release) = 4 notes
Glockenspiel (medium release) = 8 notes due to polyphony
Vibraphone/Xylophone (medium release) = 16 notes (almost like a keyboard, but depends on the score.)
Bass Drum (long release) = 4 notes

So for percussion we need 42 notes, which sums to 176 voices + 42 voices = 218 voices
and it is not finished !

e)	Strings
Don't forget : NO DIVISI or the number of Voices will grow quickly !
1 note played.

Violins:	4 voices (you will have to choose if you use 8 strings patch or 24 strings patch). Some ADSR can be a good thing.
Alti:	4 voices ( same as Vlns)
Celli:	4 voices
Double Bass:	4 voices

So, 16 voices for the" Strings section" which is the absolute minimum but again, there is no divisi which isn't realistic ! So I will count the double amount, so that you can have divisi in the strings section. We thus get 32 voices for strings.

Let's add them all:
Small Harmony	                32 voices
Horns	                            20 voices
Keyboards&Harp	128 voices
Percussions	            42 voices
Strings section	             32 voices
total:	                 254 voices

Conclusion
Can any software give you 254 notes?
At present the answer is no, there is no software able to reproduce 254 notes (a friend of mine used 2 gigasamplers to try to be more realistic...)
Of course, this example recreates a full cluster played by all musicians at the same time. This is not common, although such situations do often occur at the end of a movement or at the end of an orchestral piece.
Let's have a look at the "Score example A" picture. This is the first bar of the "Brahms Variations" page 8. We'll count the number of simultanoues notes.
There are only 13 notes, which isn't a lot, but in stereo and with some ADSR we get 52 voices.
Is it playable by a modern sampler? Yes, of course it is. So why do I need 250 voices then? As I said before, I think 250 is a sort of maximum. Also don't forget that there are "divisi" in orchestral music. 14 Strings can all play the same note but they can also play 2, 3 or 4 different notes at the same time, i.e. a complete chord. That's one of the things that will influence the polyphony needed. It is not possible to predicted the number of voices needed in "divisi" case because it depends on the "Harmony".

This is just a short reflection on orchestral music and samplers etc.
We can continue to share ideas...

Igorbolender@noos.fr
