VCSL

[Samulis]

(04-21-2018, 05:01 AM)peastman Wrote:

Quote: That's why I always try to use 'rr' and 'vl' to indicate which are those fields, and use note names rather than MIDI note numbers.

That sounds great as a convention to use.  It's just that right now, a lot of files don't follow it.   I can work around that, but if you rename the files to follow that convention then generating SFZ files can become a much more automated, less manual process.

Just saw your commit, thanks! I will check these out right away.

For the future, I have set up a new 'sfz' branch for the main project which you may work out of or commit to instead of the master. I'd like to keep the master to have only the samples, for minimal confusion. 

---

Ok, so I noticed a few things-

- 'Rel' designates Release. These regions should be appended as a new <group> (or with a new <global>) with the opcode "trigger=release". "trigger=attack" should be placed in the <global> or each <group> of the normal sustain as well for it to work right. I included an example with the ocarina below. I may add release samples to more instruments over time, as they add an incredible amount of realism to the samples, especially when without additional reverb.

- The tenor recorder is pretty quiet. The easiest fix for this kind of thing is a volume= in the global, but unfortunately volume= is overwritten if it is included elsewhere lower down the chain, so some sort of 'additive volume offset' may be a good idea for your script, so it can add a user-entered int to the calculated volume it uses.

- While having custom volume tweaks to normalize/smooth over the samples is a good idea, it can cause inaccuracies, such as with instruments like tin whistle, harp, or piccolo, which are much higher sound pressure level in their higher registers than lower. If I were to do that, I would interpolate the change, maybe to only 50%, so that it doesn't completely overwrite any natural volume curves. Obviously this is a matter of taste/opinion, so what you do is up to you.

A few more I noticed-
- Piano is 1 octave too high (note that most instruments are C3=60=middle C, only a few, if any, are C4=60)
- Piano has an attack of .1. The maximum recommended attack for "de-clicking" is 0.005- this will remove the chance of any non-zero starting points in samples from causing a click sound. The 'Release' group may have a higher 'ampeg_attack' than 0.005, up to the length of the 'ampeg_release' of the 'normal' group (i.e. 0.5); basically making a crossfade into the release.
- The recommended maximum release length is 0.3 or 0.4. Sometimes as little as .25 or so is plenty; I've even regularly done some Maize instruments as short as .15.
- Generally it is best to "favor" the low velocities more, so they take up more of the velocity layers. I typically place any high velocity layer at ~100 or above, as that seems about a comfortable trigger point. Overall, I felt the high velocity layer was a bit too easy to trigger for the piano, for example.
- If you want to win MAJOR brownie points, you could try a velocity crossfade on the yamaha piano. I don't know how well it will work, but someone tried it out in Kontakt on one of the other pianos and it came out fantastic. Phasing is always a risk with this, but it's worth trying I think.