[LAU] creating open source sample libs. (Was: Re: Experiment One - For Sonatina Symphonic Orchestra. Music and a short review of the sample library)

Nils list at nilsgey.de
Mon Aug 13 16:05:37 UTC 2012

On Mon, 13 Aug 2012 20:17:47 +1200
Robin Paulson <robin.paulson at gmail.com> wrote:

> On 6 August 2012 11:59, Nils <list at nilsgey.de> wrote:
> > Wrote a piece, 1:30 length, to test the Sonatine Symphonic Orchestra.
> > The lib is mediocre and outdated but still the best open source one.
> perhaps if there were a community around this, and if it was
> well-publicised, we could improve the quality of it, sample by sample?
> the usual free/open infrastructure, such as a github account, to allow
> people to submit their own samples?
> -- 
> robin

There are two different things:

The meta file, sfz in this case, and the audio recordings aka. samples.

I think it is not a good idea to use a git repository for binary recordings (wave or flac), but an rsync directory without any history and diff information could be enough.
But more important: Creating samples in a remote and distributed way does not work. It needs to be all from one hand or else the sound will not match. This is one reason why Sonatina is so mediocre.

But what can and and should be improved is the sfz file, which of course can be edited through git.
What can be done with the sfz:
Loop points (hard)
Extended range (very easy)
Remove the transposition from some instruments (medium)
Probably tweaking release times (technically easy, a lot of trial and error)
Make paths linux compatible (already done by someone I can't remember but probably reads this list as well)

Improving the actual sounds may be possible as well by mixing and layering the samples itself with other instruments. 
But this is very frustrating! I search the web in regular intervals for hours but finding properly licensed instrument recordings (meant for samples) is not an easy task. Probably the Sonatina author did his best by gathering all the sf2 in existence.
There are few recordings which are in pure wave format and not connected through an .sfz or similar. 
For example I found this one here recently: http://theremin.music.uiowa.edu/MIS.html
It looked very promising. Proper recordings, 4 audio technicans, proper license, well sorted. 
I only downloaded the Violin so far. Oh boy!! What a shitload of fuck. What were they thinking? They recorded this with a 10 year old violing beginner or something like this. Unstable, scratchy, shaky playing. I really hope the other instruments are better. But the violin is already unusable.

Creation of new instrument is a huge effort and even then it has been all done already. Until you are a really good recording enginner, specialized in sample recording, and have pro level musicians, and most important: a vision how you want to create this virtual instrument the project is already doomed to the lower level of quality, only sticking out because it has an open source lisence.
This is not so true for percussive instruments (percussion itself, piano, everything with "hit and forget") but for strings it is.

I see the future of instrument technology in Phyiscal Modelling anyway. But currently there are even less people interested in developing these than recording and creating sampled instruments.
At least you don't need fancy recording hardware and rent stages and a whole orchestras for months. 
Closed source, not even a real software yet, but very impressive: http://www.youtube.com/watch?v=oOBnDsz6ZmQ  (and the other 2 videos in the channel)

There are one or two Linux softwares for creating phyiscal audio modelling but afaik they were abandoned years ago at a level far away from "realisic" instruments. 


More information about the Linux-audio-user mailing list