(04-25-2018, 04:13 PM)peastman Wrote:Quote:In the case of the harpsichords, anyone who has ever played a real harpsichord or is using it for a performance will want the releases just like in real life.
Ok, if that's really what you want!
Quote:overall the instruments seem a bit on the quiet side.
For the most part I've just left the samples at whatever volume they are. If the samples are quiet, that's what you get. But if you want I can try to standardize the volume of everything. Can you point to an instrument that you think is about the "right" volume? I'll then try to make everything roughly the same volume as that one (which is a subjective judgement, but it will at least put them all in the same ballpark).
Quote:The instruments in general have their top velocity layer mapped too low
The way I implemented layer assignment is to let you specify an exponent for the mapping curve. The default value is 0.7. For two layers, that puts the division at 78. For three layers it puts them at 59 and 96. I could change the exponent to push those up a bit if you like. For example, if I change it to 0.6 that will put the break at 84 for two layers, and at 66 and 100 for three layers. 0.5 would put the break at 90 for two layers, and at 73 and 104 for three layers.
If you activate crossfading, it uses the same breaks but as if there were one fewer layers. So with two layers, it crossfades across the entire range. With three layers it fades from the first to the second between 0 and 78, then from the second to the third between 78 and 127.
For the harpsichords that include a little of the note in the release samples, should I include rt_decay? Ideally we want the pitched part to get quieter the longer you hold the note, but the click to remain the same. But that isn't an option.
My suggestion is to have the loudest samples for each have peaks between -8 and -4 dB when turned up full. For example, if you add 10 dB of gain to Alto Recorder Staccato, it will comply with this. This is enough to be fully audible, but not so much that playing block chords will be likely to clip easily (double the instruments =/= double the dB level). There are some instruments which seem to want to be quieter, either because they are in real life, or because they are at a more sensitive region of human hearing (e.g. ~3 kHz). For this reason, I tend to balance the instruments roughly by ear, but I accept that's a bit much to ask. If every instrument has the same loudness, it puts the onus on the end user to mix properly, and noting that many freeware users barely know what instruments are (I certainly was in that case when I started), I kind of have doubts that that is a good plan of action. With that in mind, I will probably go and look through opcodes in case there's something we can put in the group that affects the volume of everything...
Regardless, with SFZ, you almost can't have too much volume (well, up to -0.1 dB), as there are numerous trivial ways of reducing volume as an end user, but almost no ways (except for using a gain knob or compressor or something) to add gain.
I went ahead and made the changes to the harpsichords in the main SFZ branch. I might add back the rt_decay for the Italian, but I'm not quite sure. Many people liked the original Maize version which made no effort to mask those releases, even if they do sound a little unnatural. I had great luck getting a much more realistic sound out of the harpsis by increasing the ampeg_release of the trigger=attack group.
By the way, for some reason, ampeg_release doesn't work on trigger=release samples, so I just commented those out. I think there's a different opcode for release decays.
Sample library developer, composer, and amateur organologist at Versilian Studios.