What sort of samples you need to record for a good flute sample pack

Question

I would like to create a sample pack of the Native American flute. I am new to this, but having looked around, you don't need thousands of samples in a sample pack, just maybe 100.

What I'm wondering is how to properly record samples for a flute?

That is, what angles and such do you need to record to capture all that the flute can do? (Not talking about which mics you use and all of that, just of the tones and effects you can do with the flute).

For example:

1. Slowly turning on a note.
2. Abruptly turning on a note.
3. The screechy sound of the flute.
4. The regular sound of the flute.
5. Transitions between notes?
6. Every possible note?

For example, do you need to record every possible note you can make on the flute, or can you just record everything in "A" and then use the effect tools in your software to modulate the sound to generate the rest of the notes (and would it sound accurate).

These sorts of things. Basically, what sounds go into a good flute library, and how many (order of magnitude) would be a good set.

Basically, how much variations of samples do you collect, vs. how much can software do the rest.

Answer 1

This boils down to "How much detail do you require, how much articulation & can your playback engine support that articulation?"

At one end of the scale we have the single sample, stretched the full length of the keyboard.
This is sampling circa 1985.
To that you could add a variable velocity-sensitive attack & a filter so loud notes are brighter than quiet.

In the middle, we have

• Multiple volumes per note, which can be mapped from 1-127. This, by itself, will cover your varying attacks reasonably well. You need to decide how many layers to form this from - 2, 4, 8, 32... the 'good stuff' uses a lot.

• Multiple samples at each of your previous 'velocities'. These you then use as a round-robin so it doesn't sound like a machine-gun if someone plays the same note over & over. The human ear is very good at spotting 'the same exact sound again'.

& out the topside of realism

• partial waves which can be cross-faded so any single note can be changed over time; eg start soft then get louder - this only applies to wind (& bowed) instruments, but it's a great deal of the 'reality' of how they work. This would include overblow [otherwise you'd have to fake that into the top 10% of your 'velocity set'].

• Algorithmic legato between notes, or samples of every possible legato jump at every velocity.

• 'Noises' - breaths, finger noise, taps & releases. These are all subliminal but contribute to the overall realism.

Additionally, the more detail you require in your data, the better a player you need to generate it & the better a sound engineer you need to record, edit, map & collate it.
You will need a good set of naming conventions too, so you don't get lost in the data pile.

https://audiomodeling.com has some great examples of the 'topside' of realism & there's a description of how their SWAM-W Engine works at https://audiomodeling.com/technology/swam-w/