Faking sections from solo instruments (a.k.a. "Four for the price of one")

[Mattias Westlund]

Following this discussion I decided to create a small violin section using ld1609's samples as an example of what you can do if you have a chromatically sampled solo instrument and want to fake a section sound. I realize that the sfz files themselves aren't entirely clear at a glance so let me add some brief notes on what is going on in them.

I have created three different versions of the same instrument by mapping it in minor thirds instead of chromatically. First violin starts its mapping from G, the second on G#, and the third on A. Like this:

Violin 1: G, A#, C#, E, G (etc)

Violin 2: G#, B, D, F, G# (etc)

Violin 3: A, C, D#, F#, A (etc)

The first and last samples are then stretched up and down so that all three ranges match each other.

As I felt three instruments weren't quite enough, I added a fourth one identical to the first except for that I detuned it by 6 cents (tune=-6) and added a 20 ms delay to its note start (delay=0.02). I panned the first and fourth more to the sides as to not make them clash, and kept the other two violins closer to the center. For this I used the position opcode.

For the staccato and pizzicato articulations I also created fake 2xRR by duplicating the existing sample layers and messing around with the offset and ampeg_attack values until I had something that sounded like variations of the same note.

Obviously the end result does not sound exactly like an actual four piece violin section. But it's not half bad either and with a bit of more work, like for example randomizing certain parameters like attack, release and note start by small amounts, you could make it even more convincing. This I leave to the SFZ gurus out there though, because working in this format gives me headache

[Mattias Westlund]

As an interesting side note, I tried to do the same with the MSLP cello but this didn't work even nearly as well as the violin samples. It turned out all buzzy and phasey sounding. Which goes to show that the success rate of this method largely depends on the character of the samples you're using. Some can happily be layered, some can not.

[Samulis]

Probably a big factor that might help with that is stereo positioning. I do believe SFZ has stereo positioning controls somewhere in the opcode spec...

Ah, it is called Position-

Quote:Only operational for stereo samples, position defines the position in the stereo field of a stereo signal, after channel mixing as defined in the width opcode.

A value of zero means centered, negative values move the panoramic to the left, positive to the right.

Examples:
// mix both channels and play the result at left
width=0 position=-100

// make the stereo image narrower and play it
// slightly right
width=50 position=30

You can also use some slight random detuning (maybe 12?), random delay (0.1-0.2?), and random offset on each sample-

Quote:Random tuning for the region, in cents. Random pitch will be centered, with positive and negative values.

Examples:
pitch_random=100
pitch_random=400

Region random delay time, in seconds.
If the region receives a note-off message before delay time, the region won't play.

Examples:
delay_random=1
delay_random=0.2

Random offset added to the region offset, in sample units.
Examples:
offset_random=300
offset_random=100

(btw, this random detuning and delay is an effect used in GPO mapped to a MIDI CC... 32 & 33 I think? Anyway, it's a really awesome effect I think I am going to include in my own sfz build soon)

You can really go pretty crazy here with rand stuff to get what you want. Theoretically, if you put on a decent attack length and just rand generate sample starts with a big enough range and long samples, you should never need to worry about not having enough RR's (even better if you have separately-cut attack/transients that are layered on top with a random filter and random cycling as well).