How to simulate performance volume levels in MIDI playback

Question

I've a questions about preparing a brass band arrangement in Sibelius, although the answer would also apply preparing parts in a MIDI sequencer and then creating notation from them.

In a UK brass band score, the top five staves are performed by the cornet section. In the cornet section we have:

• 1 x soprano cornet
• 4 x solo cornets (playing the same line),
• 1 x repiano cornet (like another solo cornet, but there's only one player on this part)
• 2 x second cornets
• 2 x third cornets

Once I've written five staves of music and ask Sibelius to perform those five staves, all the lines play back at the same volume and it all sounds balanced. But - if I put the same notation in front of actual players the balance will be completely different because some staves have more players.

I know that two players isn't twice as loud as one player. Not really sure what the rules are.

So my question is this: how do I get the playback of the five staves to accurately represent the balance I'd get if the score were to be performed by an actual cornet section?

To clarify, the end product is a printed score which will be performed by actual musicians, not some ghastly faux brass band audio track. I'm not really bothered with the sound quality from Sibelius, I'm using it to check that my harmonies are OK. I'm actually using Sibelius with GM sounds rather than samples. My question really is: what adjustments should I make to the Sibelius playback to give me a better idea of the balance from the actual musicians when it's actually performed?

Answer 1

Don't over-think this. Live players listen, adjust and balance. They don't interpret 'f' as a fixed dB level, 'p' as another. They make it work.

A sequencer will only give you a very general idea of how live players will balance. Your practical experience of bands and knowledge of scoring practices are much more important.

Answer 2

Consider playing to make a score & playing to make it sound like a section two totally separate jobs.

What most people try to do when generating large sections in Midi is they try to play it all at once like a pianist, or worse, they try to generate it from a score they clicked in with the mouse. This just never works, and you'll spend more time trying to fix it than if you just played it all again.

Play each line one at a time. Ensure each performance sounds & feels like a performance. Layer them up. Needless to say, you never quantise anything.

This way, not only do you get the opportunity to compensate gross levels between the instruments, as each is on a separate midi track, you also will never get all attacks, volumes, glisses, bends, vibratos etc in strict time with each other. There will be constant natural variation.

My common mappings for such would be to have mod wheel mapped to volume (velocity to volume is useless for brass & strings), PB as is, velocity to attack & AT to either mod/vib or overblow (or similar), or I'd run a second pass using an iPhone running TouchOSC (there are quite a few variations on this these days if you look at the App Store) with a page of modulation controllers I could then overdub in pseudo-realtime. This was useful if mod/vib needed to itself have speed as well as intensity variation.
A lot of these decisions would depend on the capabilities of that particular sample set.

Another trick is don't use all the same sample set to make up a section, use two or three, even from different manufacturers. I would always lead with my most expressive set & use the more generic stuff which required less manual controller work for the lesser parts, or 'second fiddles'.

Answer 3

I agree with Laurence suggestion to do it by ear. If you don't, you'll run in many kinds of trouble. I'll try to list them:

• 2 instruments sound 3 dB louder than one; 3 instruments sound 5 dB louder than one; 4 instruments sound 6 dB than one. So adjusting volume of the track by several dB might work...
• MIDI velocity values (which you can adjust e.g. using dynamic marks like piano, forte) are supposed to reflect the instrument response to musician dynamics. This includes loudness, but also sound characteristic – in many instruments loud notes sound also brighter. Moreover various virtual instruments may respond differently to note velocity (this is called velocity curve). So that's not a good way to adjust balance between playback of virtual instruments.
• Ensemble of instruments sounds differently from a single instrument. To emulate that you'd need a virtual instrument like "cornet section", rather than "cornet solo". You could try to emulate this by adding a light chorus effect, however note, some chorus effect might affect signal volume as well
• All of the above assumes that the volume of the virtual instruments you use is normalized to match differences between loudness of real instruments. Is it? Quite likely not. Rather the volumes are adjusted so that all instruments have more less the same loudness.

Answer 4

I agree with the given ideas adjusting by ear, chorus effect.

But what is your goal? A demo tape? Midi sound cards will never compensate original sounds. Brass instruments in midi are horrible any way. If you want a proper recording of your piece let it play by a band.