This is the first time I’m testing the new two piano sfz accuratesalamander and j3drhodes, because I finally installed the new beta sd image. Before that, I was using the full stereo Rhodes and salamander grand piano with linuxsampler (sorry to have to nag you again with this one). That was with vangelis but without upgrading the SD image.
The problem now:
The new sfz versions of these two piano sounds do not produce any sound with linuxsampler, but they do work with sfizz.
Is this expected? Does anybody confirm this?
The reason why I preferred the situation before (salamander grand piano and stereo Rhodes with linuxsampler) is that I have the impression that it produced better quality (no idea why… resampling quality? Dynamics? Different touch sensitivity? Maybe it’s just in my head / ears).
Also sfizz sporadically cut off notes when it shouldn’t, but I haven’t had this problem since the upgrade of the SD image.
By the way, the upgrade process went like a charm for me on my stock 5.1, big thank you!
Hi @emuse! Thanks for the positive feedback. We are moving away from LinuxSampler for many reasons, not least because it is proving challenging to maintain two SFZ engines and keep both working well (as demonstrated by your report). Our first answer will increasing be, “Please try sfizz” because we want to concentrate on making this as good as possible, fixing issues and engaging upstream where necessary. If there is no problem using sfizz then we won’t start to look at LinuxSampler. (We may be unlikely to look anyway, preferring some workaround or acceptance of limitations with sfizz!)
Please try sfizz again and if you feel it suboptimal, try to quantify issues so that we can look at resolving them within the sfizz code (which is truely open source, unlike LinuxSampler with its weird and sometimes unfathomable clauses).
These two use flac files. It might be worth to check if the current linuxsampler version was compiled for usage with flac samples.
For focussing on sfizz I’m completely with @riban. Sudden note off problems could be rooted in the polyphony limit when too many samples are loaded per note-on (different mic positions, velocity fades…). E.g. try the 3d rhodes with the respective non crossfading patches (keyswitch Ab7, Bb7), these should use est. half the samples per note on.
Hi, thanks to all of you for your thoughts. I believe now that the sound differences I hear might come from the sample sets themselves. I would like to check this, and also the sfizz note cut-offs I had earlier (doesn’t happen with those new two sfz pianos). Is there still a way to install the original versions of stereo Rhodes and salamander grand piano? Those two had a larger sample set if I remember right. This way I could check how they play along with sfizz, and the sound quality at once.
I think it might be good for the “factory library” to chose the best versions for each kind of instrument. Besides that, the stereo rhodes had an uncanny license question, and the new Salamander really performs better. If you want to use the original files you might want to throw them in your /zynthian/zynthian-my-data/soundfonts/sfz folder after downloading from these sources:
As far as I understand: someone presumably vibecoded MPE support in a (probably nonfunctioning) sfizz fork, which is now independent fork named sfizioso(-player) after sfizz maintainer rejected reviewing 3000 presumably vibecoded lines and archived sfizz. Someone else again created another fork of sfizz called sfz-ui-mpe using the mpe code above.
Since @jlearman might heard more about it, can you tell something about it?
The archival of sfizz was discussed a bit on the sfz or sfz-tools discourse, but not much more info than already reported above. I believe that folks are trying to get sfizz back in business, without thousands of vibe-coded lines – those belong in forks.
linuxSampler isn’t a very complete or up-to-date sfz player, and I don’t test my sampled instruments with it.
LinuxSampler does support FLAC, I believe, so it’s probably something else.
Regarding the changes to factor samplesets, I think it’d be a good idea to have an “Oram factorysoundfonts” package. The problem with this is that it overlaps the Vangelis ones, so uninstall could have unintended consequences. But we could manage that with documentation: When both are installed and one is removed, re-install the other. IMHO that’s simple enough for the user and wouldn’t make it complicated to produce the Oram sampleset package.
I tested the Salamander Grand Piano from the link you gave @hannesmenzel thank you! There is quite a big difference between that and the Zynthian Grand Piano installed from the package. Not criticizing the hard work to provide that latter one (which sounds nice and can do the job probably for most use cases), but I just prefer the older, big one for playing along with the headphone. It’s all subjective isn’t it…I had no problems with sfizz since then even with the big original sample set, and I’m happy with the sound, too.
As to the j3drhodes, it sounds really good to me indeed !
So the sound “problem” was not related to sfizz vs linuxsampler, and I can let go the latter, too. Won’t nag you anymore with it , sfizz seems to work flawlessly here now, who knows what was wrong on my installation before the new sd image…maybe access time of the samples on the earlier memory location on the sd card?
The “newer one” (Accurate Salamander) also has a “bigger one” version which is also available as a package. The smaller one is @jlearmans version for reduced footprint.
I didn’t know this. How can I find this package? I can only see “Accurate Salamander Grand” in the packages on webconf, and that installs “Zynthian Salamander Grand” as a preset for sfizz.
Edit: or do you mean on github directly, not in the zynthian webconf? Thanks!
Yes, right this. If I’m not completely mistaken the currently preinstalled is a reduced version of the accurate Salamander (currently version 6 or so) from here:
And the extra package from webconf is the non reduced one.
Your favorite one you mentioned above is the old/gold V3 from the github link above.
Are you willing to try a different version? The reduced version I created earlier did the following to reduce footprint (and in some cases voice usage) :
16 bit samples rather than 24 bits – believe me nobody can hear the difference
FLAC (lossless compression) vs WAV file format
deleted pedal-down samples
deleted release samples (this also reduces voice count)
The first two items don’t change the sound at all (except, it’s based on a later version of Accurate Salamander Grand, so there might be issues there but I doubt it – IMHO each release of that project is an improvement with nothing lost.)
My guess is that replacing the pedal-down samples would make it sound a lot nicer. When I did it I was hoping that my piano resonance plugin would provide that instead, but that’s too fiddly, and it does use a bit of CPU.
But it might also be the release samples (the sound of the hammers falling back when releasing a key.) These don’t take much disk space, but they do consume voices, and especially on RPi4 the voice count limit is severe.
Please LMK if you’re interested in trying, or any feedback on which of the above two matter to you.
I’d like to reduce the footprint without reducing the experience.
Also, LMK if you’d like me to walk you through trying PianoRes with the sampleset you didn’t like earlier (we can have both loaded rather than replacing what you like.) You might like having direct control over the amount of harp resonance.
Thanks for your explanations. Yes of course I’d like to test an “intermediate size” version. Yes I think the hammer noises are what I missed in the zynthian grand. The full version plays well on a 5.1, at least when there is not much other stuff playing at the same time. The drawback I would see is some seconds of loading time when switching preset between two zs3 states for instance. So it makes perfect sense to reduce footprint. But I think the full version should remain available as a package alongside… as we have it now.
, sfizz seems to work flawlessly here now, who knows what was wrong on my installation before the new sd image…maybe access time of the samples on the earlier memory location on the sd card?