Re: [fluid-dev] Overflow settings (was: problems with fluidsynth 1.1.6 on a raspberry pi)

[Aere Greenway]

David:

Thanks for the additional information, and enlightenment on aspects of the problem I hadn't thought of.  

As a software developer, I have learned from long experience that the internal workings of a software package are much more complex than will be realized by an outside observer.  

I will experiment with the settings, but this effort will be time-shared with final-development-phase work on my own MIDI creation, so the results may be a bit slow in coming.  

-- 
Thanks,
Aere

-----Original Message-----
From: David Henningsson <address@hidden>
To: Aere Greenway <address@hidden>
Subject: Overflow settings (was: problems with fluidsynth 1.1.6 on a raspberry pi)
Date: Mon, 17 Dec 2012 08:32:35 +0100

On 12/17/2012 12:45 AM, Aere Greenway wrote:
> David:
>
> Establishing a score on a note to compare with other notes' scores, to
> determine which note to remove is an interesting idea.

I was a little inaccurate in my previous email; the score is per voice, 
i e, if a note translates to more than voice in the soundfont, all 
voices are scored independently.

Anyway it's the method used since before my time with the project.

> But there might be an easier way to determine whether a note can be
> killed or not.  I am merely speculating here, and don't know what
> information is visible to the code that decides to remove a note.
>
> Many instruments have a sound that fades-away with time, regardless
> whether or not the sustain-pedal is pressed.  The volume it fades-to is
> probably what is maintained sounding by the sustain-pedal.
>
> Although percussion instruments (in a rhythm-track) are important for
> keeping the other instruments together, many percussion instruments
> (such as a wood-block) are the type whose sound quickly fades to a
> barely audible (or inaudible) level.  Also, the sustain pedal doesn't
> seem to affect them.

When a sample has finished playing it is automatically freed up so that 
it can be used for other voices. A wood-block sound would typically have 
a very short sample.

> Likewise, a guitar sound has a strong attack sound, but quickly fades to
> a lower level.
>
> Playing a guitar voice with the sustain-pedal held down changes the
> sound, making it generally richer and more full, but it doesn't have as
> noticeable an effect as playing a piano voice with the sustain-pedal
> held down.
>
> On the other hand, very many instruments (such as on oboe, trumpet, or
> French horn, to name a few), have a high volume-level for the sustained
> note (which is lower than the attack-phase of the note).  In
> performing/recording a track for such instruments, I generally don't use
> the sustain pedal, but will use it if I need to keep a note playing
> while I re-position my hand to play the next phrase.
>
> If the software is aware of the volume level of notes still sounding, it
> would seem logical (to me) that if a note is no longer sounding at an
> audible (or barely audible) level, the note could be removed without
> making a noticeable change to the sound.
>
> So it makes sense that the volume level of a note is a very important
> (perhaps most-important) part of the score used to determine if a note
> could be removed.  Certainly if a note is no longer audible (even though
> the sustain pedal is depressed) it could be removed with impunity.

Sure; but things are more complex than that. We also have stuff like 
reverse cymbal, where the note is currently almost inaudible, but soon 
will have a very strong sound.

Or imagine that somebody has very soft sound, maybe a synth pad or so, 
that seems like a good candidate for removal, but later the user want to 
use the channel's main volume to achieve a techno-like effect on that sound.

So; I'm not saying there isn't room for improvement, the current 
algorithm is far from optimal. But there is also another constraint - if 
we were to do some kind of RMS calculations of every sample to find its 
maximum volume (or even future maximum volume?) at different points in 
time, that is time/CPU consuming. I'm not exactly sure when to do that 
to make it accurate without sacrificing latency at the same time. I e, 
determining how loud a voice currently is, or will be, is far from trivial.

> I normally use Qsynth, and am thinking there is no way (at least in the
> GUI) to change the parameters.
>
> In looking at the man-page for Qsynth, I saw the following option info:
>
>         -o, --option [name=value]
>                Define a setting name=value

Ok, that is good to know!

// David

-- 

Sincerely,
AereSincerely,
Aere

-- 

Sincerely,
AereSincerely,
Aere