Re: [fluid-dev] audio processing stops on ubuntu karmic

[josh]

Hello,

Quoting Joshua Higgins <address@hidden>:
> I'm using fluidsynth 1.0.9 (packaged with karmic) on an old lifebook
> b-series laptop. I have a very large Studiologic SL-880 midi controller, so
> i'm intergrating this laptop into the body of the keyboard, using fluidsynth
> for sound.
> The PC is low spec: 500mhz with 192mb ram. I've bootstrapped Ubuntu Karmic
> on a 4gb CF card which fluidsynth is running off. Its using the realtime
> kernel, with jackd running with the --realtime parameter.
>
> For normal playing, the performance is excellent. Jack seems to be stable,
> running with 512 frames and 2 periods yeilding a latency of 10.7ms.
>
> However, if I press down lots of keys simulatenously, i.e. more than 3
> octaves at once, the audio stops straight away with the error message on the
> console:
>
>
> subgraph starting at fluidsynth timed out (subgraph_wait_fd=9, status = 0,
> state = Running, pollret = 0 revents = 0x0)
> bad status (1) for client fluidsynth handing event (type = 8)
>
>
> The strange thing is that both the jack server and fluidsynth are still
> running, just audio processing has stopped.
>
> Can anyone shed any light on the situation? Is this computer just too slow?
>
> Thanks.

My guess is that FluidSynth may be consuming all CPU and Jack is  
therefore timing out waiting for it to finish an audio block and  
subsequently disconnecting from FluidSynth.  You probably want to set  
synth.polyphony to a value that that system can handle.  This is  
probably going to be something rather low for that system, like 32 (3  
octaves would be 36 keys).  I'd start small, play a bunch of keys, see  
where the CPU usage is at, then increase it, until you find the point  
at which it breaks.  You may also want to increase the Jack timeout  
value ("jackd -t 2000" for example would set it to 2 seconds).   
Turning off Reverb and/or Chorus might save you a little CPU usage as  
well, if you can do without one or the other.

Note that once the max polyphony value is reached, voice stealing will  
occur, where "less audible" voices will be turned off to make room for  
new ones.  This logic currently needs some improvement in FluidSynth,  
since sometimes voices are turned off which aren't ideal.

> --
> joshua higgins
> > > > > > > ------

Hope that helps!

Josh Green