Re: Re: [fluid-dev] FluidSynth 1.1.2 release planning

[David Henningsson]

> Hi David,

> - rendering: 3. But no distortions because of audio level etc. The
> recording is taken directly from live playing a real keyboard, using a SB
> Live SB0060 (Platinum) and the built-in EMU 10K1 EAX reverb (native
> Creative driver), through WinXP WDM DirectX dSound driver. Since this is a
> real-time instrument, no other strategy is useful.

Could you try file rendering with a bigger buffer, and see if that makes a
difference? That corresponds to "-a file -z 4096" in FluidSynth
executable. You won't hear what you play, but it'll still be real-time and
you can listen to it afterwards.

> - polyphony: I just explained our usage of the FS polyphony offer. The
> main issue is that I have to cancel one preset for starting another, also
> while using just one preset holding just one key pressed (a behaviour I
> know also from the Creative synth, due to it's limited memory). I've no
> explanation for this, because all parametres in FS are set correctly so
> far. It was the same already in FS 1.0.9.

There can only be one preset per channel at the same time, that's a design
decision.

> Did voice stealing work at all,
> in 1.0.9 and 1.1.1 releases?

Yes.

> - system: after all, I cannot provide rock-solid informations. But to me
> it seems, that a big Level2 cache of the processor is the main point for
> lowering latency.

Interesting! Perhaps I should run it through cachegrind and see if there
are any obvious things one could speed up.

// David