Re: [fluid-dev] Delay and buffer size

[Josh Green]

On Thu, 2006-10-12 at 19:41 -0700, Z F wrote:
> Hello everybody,
> 
> I have read in the list archives that if there is noise while switching
> windows, than the buffers and the number of them needs to be increased.
> this will, however, will increase the delay in responce.
> On my system I had to do fluidsynth -c 16 -z 1024 to get rid of the
> noise. The question is than is, what is the delay in ms(?) compared to
> the default. And also, what kind of delay will not be noticable?
> 
> Thanks for your help
> 
> ZF
> 

Hello.

It depends on what sampling rate you are running at.  But if you are
running at 44100 then you should calculate the delay like so:

(16 * 1024) / 44100 = ~372 Milliseconds

For live work, that would likely be very noticeable latency.  You can
likely get better latency (with little or no artifacts) by running
FluidSynth as root.  There is currently the danger of machine lockup
though, if all CPU is consumed.  This can be limited by setting the
maximum number of allowable voices.  There may be latency issues with
the kernel you are running too.  That's an art in itself though, in
getting a good low latency kernel.  Some kernel versions in 2.6 are
rather bad, others are better.

One other thing to try is make sure you are using the ALSA hardware
device rather than the plugin system.  This can be done by passing the
following switches to FluidSynth "-o audio.alsa.device=hw:0".

Max number of voices can be set with: "-o synth.polyphony 64".  That
will set the max number of voices to 64.  You may want to try something
lower perhaps first when running as root to make sure all your CPU isn't
consumed when all voices are used (a very good possibility when playing
MIDI files).

There is likely a lot that could be done coding wise with FluidSynth to
improve things, in this regard.  Not sure when I will get around to it
though.
        Josh Green