[Freepats] multilayer editing with swami

[Roberto Gordo Saez]

On Sun, Jul 01, 2007 at 04:57:18PM +0200, Marcos Guglielmetti wrote:
> Feel free to write me in english: I can understand it perfectly.
> My problems is that sometimes I dont have time to write in english.

OK.

> BTW: you did not made any loop to the samples into the sf2 file, why? I mean, 
> I can understand tha we like HQ, but... from the 70% of the sound sample and 
> up to the end you almost cant listen.

Maybe you have some backgound noise on the room? I have a passive cooled
computer and on a silent room I can perfectly hear the sample to the end.
Try with a chord instead of a single note, I'm sure you will also be
able to hear full samples at a normal volume.

Signal/noise ratio is very poor in the softer part, though. Using sorter
samples will help to avoid the noisy parts.

> I can understand that this is the best for a HQ sf2, but in the future we 
> could use some interesting loop methods, using "until release"  loop mode, 
> and merging parts of the sample in the way tha we could make perfect loops, I 
> will show you something like this if I could do it.

Correct, loops are missing. Take a look at this mail by myself:

http://opensrc.org/pipermail/freepats/2007/000117.html

Choosing good loop points (in a way that they can't be noticed) takes
many time, and I'm lazy, so that is the main reason for no loops :-)

> Also, the IMIS steinway sound samples are not normalized, I can also 
> understand that the pp and mf layers must not sound with peaks of 0 db, but 
> the FF layer I think it could do it.
> 
> The problem with this issue is that in that case we will have to increase the 
> overall volume, not just the FF layer... I will see.
>
> For instance, FF b2 needs 11.9 db to be at full volume, this will allow us to 
> listen to the final sound at the end of the sound sample 

I've tried to use a tool called normalize-audio:

http://normalize.nongnu.org/

The result was not so good. I've made some research, and it seems that
an acoustic (non electric) piano is a very difficult instrument to
balance, because all parts (attack, sustain...) may require different
gain levels. It looks that some professional (and expensive) piano
soundfounts are normalized by applying a custom amplitude envelope to
each individual sample, balanced by ear in comparison with all the other
notes. That's a lot of work!

I've opted to just apply a certain gain to each sample and forgot about
this issue for now. It is far from perfect. You can find the samples
I've used here:

http://freepats.opensrc.org/sf2/acoustic_piano_imis_samples/

They are cropped, but before applying the gain. I think you could start
from scratch anyway, since very little is actually done. Now you can
understand why I've said the soundfont was only a demo, eh? :-)

Well... just try to make something better each time, and with enough
time we will end with a very good soundfont.

> what do you think about this and  the other issues?

Some samples contain "clicks" or other noises inside, try to avoid them
(look for my previous mails). Also, let me suggest you to keep a copy of
the processed samples used by you, but before they are normalized. They
would be useful if at a latter time we want to "renormalize" them in
other way.