Scoring Central

(04-24-2018, 04:59 AM)peastman Wrote: [ -> ]I'm finding the releases work much better on some instruments than others.  On the French harpsichord, for example, they sound fine.  They're just very brief clicks that add a bit of realism to the end of the note.  On others, like the Italian harpsichord and Kawai grand piano, they're terrible.  When you release the note, you get an echo a moment later.  With the Kawai, that's because they're quite loud and, in some cases, quite long.  With the Italian harpsichord, it's because each release sample really does contain an echo.  There's a bit of pitch, and then a click a moment later.  So maybe that's how it's supposed to sound, but if so, it sounds really bad.

What should I be doing about these?  Should I reduce the volume of the release samples?  Add an offset to skip the beginning of them?  Just not use them?

Then there's the Yamaha piano where the release samples are all completely silent.  Clearly something went wrong with those.

Italian Harpsi is exactly as it should be. My advice is to make sure you are using a release on the normal group which is equal to the attack of the release group. I suggest 0.1 as a good value to start with for this. The same is suggested for the Kawai pianos.

The difference between Italian Harpsi and French Harpsi releases is that Italian Harpsi releases were cut from actual 'premature' ends of notes, while on French, they're just the sound of the key action with no pitch cutoff. In the former case, a short and equal release/attack as suggested above is best. In the latter, having a bit longer of a release (up to 0.3 or even 0.4) on the 'normal' and a quick attack (0.05 or less) on the 'release' is best, as that basically simulates the dampening of the note (if you wanted to get super accurate, you could put a highpass on it with an envelope to simulate the dampening action, of felt on metal strings, but I that is far well and beyond the scope of anything we're doing here).

The pianos and the Italian Harpsi might benefit from some 'amp_veltrack', maybe somewhere between 50-100 (I know, I know, I told you to put it at a low value... that really only should apply to pitch-less, non-normalized samples like in French Harpsi). This should definitely be done if you're finding it sounds atrocious when playing quiet notes, but okay with louder ones.

The Yamaha release samples aren't silent, they're just VERY quiet. Try adding 30 dB of gain to all the release samples or so and see what happens. Release samples were never planned, so those had to be dug out of the recordings- it's almost entirely just mechanism/key thumps, which isn't very attractive even when you can hear it. 

Oh, I just remembered... try sticking this into the release group for the Kawai piano/pianos in general (doesn't quite work for harpsichords, which have a release which is primarily the mechanical click). It will slowly reduce the volume of the release samples so if you hold the note out a short time, you get a loud release, but a long time, a quiet one. Will definitely need some tweaking; I don't know the proper number in the slightest. Bigger # = faster fall off/decay; smaller = slower fall off.

Thanks, that helps a lot!  I managed to get reasonable results for most of them, mainly by just bringing the volume of the releases way down.  rt_decay also helps.

I think the first batch of instruments is ready to merge.  It includes most of the pitched instruments.  I'll start working on supporting percussion instruments next.

(04-25-2018, 03:34 AM)peastman Wrote: [ -> ]Thanks, that helps a lot!  I managed to get reasonable results for most of them, mainly by just bringing the volume of the releases way down.  rt_decay also helps.

I think the first batch of instruments is ready to merge.  It includes most of the pitched instruments.  I'll start working on supporting percussion instruments next.

Looks great so far! A few little observations with the set so far... some of them are more for my/our collective sake as we go to hand finish these, but others may be able to be acted on in the scripting/generation side of things in advance-

1. The staccatos have too quick of a release, but this is very easy to fix by hand, so don't worry about it.
   
2. The Concert Harp & Folk Harp xfade doesn't seem to work all that well. Perhaps try doing a smaller crossfade zone, perhaps from 70 to 100? The lower velocity isn't appreciable at all at present, at least in my opinion... other option would be to see if adding gain to the lower layer does anything. Either way, harps tend to stay in the middle dynamic a lot and bring in the upper dynamic only once in a while to emphasize a chord, so I typically tuck the top velocity of the harp up above 100 and don't crossfade to it, so the player can just 'goose' the note and get that hard pluck, but play the instrument normally without accidentally activating that upper dynamic (which is normally avoided). This is probably something you want to address in the scripting part.
   
3. The Strumstick should not have the three strings overlaid at the same time on the same note- it creates some rather unpleasant out-of-tune bits. My recommendation is to map string 3 (D'), then further down on string 2 (A), then the rest down on string 1. This way String 3 (the most important string on the instrument) gets most of the action. This can be done by hand though.
   
4. Either Kawai - Legacy is really loud or (new) Kawai is very quiet. Probably best adjusted in script?
   
5. I would suggest not making any reduction in the volume of the Flemish, English, French, and 'Unk' harpsichord releases. The volume they were recorded at relative to their key press is the volume they should be played back at- rt_decay should not be used on a harpsichord as a result; i.e. yes, they're really THAT loud. 
   
6. Harpsichords should all have "amp_veltrack=0" or a value close to 0, as the instrument has no real dynamics. Likewise, recorders and flutes can have a reduced veltrack.
   
7. EDIT: Flemish Harpsi's 'High' should be 8vb so that if layered on top of 'Low' they sound in octaves.
   

As for what to do next, there are quite a few pitched idiophones (marimba, vibes, glockenspiel, etc.) which will be a walk in the park to map, although some of them (I'm thinking maybe Xylophone and the super legacy Tubular Bells in .ogg format) may use the old C4=60, so check ranges with those against an existing library or an orchestration range chart and let me know what is messed up. There's also THREE sets of tubular bells just waiting to be mapped (well, all technically of the same exact physical set, just different samplings). I will have a few more idiophones going up as well this week (Balafon and some types of lamellophones {mbira, kalimba, thumb pianos, etc.}).

If you really want, I can make a version of the harpsichords with the releases at full volume.  But frankly, I can't imagine anyone ever wanting to use them.  I've attached the Italian harpsichord so you can hear what it sounds like.  There's always a balance between trying to recreate a real instrument as accurately as possible, and trying to create the best possible electronic instrument for people to use in their own music.  I lean more toward the latter than the former.

Quote:Either Kawai - Legacy is really loud or (new) Kawai is very quiet.

 
The legacy one is really loud.  I can bring it down a bit.  And I can get the other fixes to you tomorrow too.

(04-25-2018, 05:10 AM)peastman Wrote: [ -> ]If you really want, I can make a version of the harpsichords with the releases at full volume.  But frankly, I can't imagine anyone ever wanting to use them.  I've attached the Italian harpsichord so you can hear what it sounds like.  There's always a balance between trying to recreate a real instrument as accurately as possible, and trying to create the best possible electronic instrument for people to use in their own music.  I lean more toward the latter than the former.

Quote:Either Kawai - Legacy is really loud or (new) Kawai is very quiet.

 
The legacy one is really loud.  I can bring it down a bit.  And I can get the other fixes to you tomorrow too.

Just pushed the new samples! I just need to do a bit of general file renaming and then I will push a 0.2.2 update that will be good for you to work off of. Hopefully by 0.2.5 all the file renaming will be done.

• Added Wine Glasses, Kalimba (Tanzania), Kalimba (Kenya), mbira dzaVadzimu Nyamaropa, mbira Mavembe/Gadanga, Nyunga Nyunga, Balafon, and Xylophone soft and hard mallets.
  

EDIT: I do want to note that the plucked idiophones (lamellophones)- kalimbas, mbiras, nyunga nyunga, etc. all have the keys listed (k1, k2, k3, etc.). It is not uncommon to map two keys to the same note, but on opposite sides of the instrument. While these can be treated as RR for a chromatic interpretation, the key numbers can be used to map the original tuning/scale of the instrument to the keyboard along, say, the white keys (e.g. k1=36, k2=38, k3=40, k4=41, etc.) (note that the tuning is not necessarily western, and while I do not suggest tuning it, I leave that decision up to you).

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. Actually, I can't really imagine a case where someone would want the releases to not sound right except perhaps for synthy electronic music... in which case sawtooth waves or whatever are the way to go. 

This is the English harpsi sample set (virtual) with no modifications, i.e. what it sounds like in real life-
https://instaud.io/PUD

A live example on another (forgive ol' fumbly-keys here). That "thump" at the end is just what harpsichords do in real life-
https://instaud.io/VU8

If it really bugs someone, then they can always go and turn off the release layer. That's my philosophy with this freeware- just get it sounding as close to the real thing as I can, and if someone wants to come in and process the samples or do whatever, then that's up to them. That's why all the samples are right there.

There are some instruments out there which are "acquired tastes" and take time to develop an appreciation for, often because they act or sound alien compared to other instruments. Many folk/non-western instruments have very complex overtones or use buzzing/rattling sounds which appear very odd to us, used to our pure tone instruments, but after listening to a lot of those sorts of instruments lately, I have come to appreciate that buzzing and complexity. One of the harpsichords has releases almost as loud as the attacks, but I have come to appreciate that behavior and it doesn't bother me in the slightest.

---

One quick observation I've had... overall the instruments seem a bit on the quiet side. I normally listen to music while I work on the sample managing stuff (nothing too compressed), and every time I want to play the instruments, I have to pause the music, turn up my headphones a fair bit, then play the instrument- even with full volume in the ARIA mixer. I think the script is lowballing the volume setting a bit.

Doodling around with English Harpsi, I managed to get a good sound with ampeg_release in the normal at 0.35 (since the release has no actual pitch content, it needs a bit of pitch content to sound right). Maybe something to change in the script? Not sure if that helps at all.

Some reflections regarding the script/automapping after more playing around and thinking-
1. The instruments in general have their top velocity layer mapped too low, so they come off as overly harsh. MIDI tracks I play through the pianos have far too much of the top velocity layer compared to how they sound on other instruments. Try the top velocity having a lovel between 90 and 100.
2. The instruments in general are on the quiet side.
3. The crossfading on the harp may work better if concentrated in a higher velocity range (e.g. 70-110).
4. I had much better luck with releases of .3 to .4 across the board. For some reason, SFZ numbers always seem to want to be a bit bigger than Maize numbers.
5. Your scripts just saved both of us about 20 years of our lives, congratulations.

I always intended to ask about your intent on velocity mapping and I don't think I ever did. I think i map high, but maybe i don't. On two velocity layer instruments I tend to do 81-127 for the high and on three velocity instruments 97-127 for high and 65-96 for mid. I've always felt that people overplay stuff so I would rather have the top 2 layers in the top half and the soft have the bottom half or the soft will never get used. I kind of figure that for most people say 24-0 will never actually be used.

Anyway just put out the VSTi for the 4 recorders all in one sort of mini-rompler. Any mac guys available to see if it works right?

http://bcvsts.blogspot.com/2018/04/vcsl-...ments.html

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.

(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.

There are a few instruments that are pitched, but the samples are labeled as if they weren't.  For example, the wine glasses and the timpani.  Even though each sample corresponds to a particular pitch, they're just labeled as "glass1", "glass2", etc.  If you like I can rename them to include notes, then process them as pitched instruments so the intermediate notes will get filled in.