Re: [fluid-dev] some notes of Cadenza.sf2 don't sound right

[Josh Green]

On Sat, 2007-03-17 at 13:44 +0200, Mihail Zenkov wrote:
> 
> I write before about this bug: it were strange and hard to reproduce -
> i can't it reproduce on my machine, but can on another. Try this very strange 
> hack:
> 
> diff -ur /mnt/doc/source/cvs/fluid/fluidsynth-ORIG/src/fluid_voice.c 
> fluidsynth-hack/src/fluid_voice.c
> --- /mnt/doc/source/cvs/fluid/fluidsynth-ORIG/src/fluid_voice.c 2007-03-04 
> 19:31:12.000000000 +0200
> +++ fluidsynth-hack/src/fluid_voice.c   2007-03-17 13:33:53.000000000 +0200
> @@ -592,6 +592,8 @@
>                     + voice->viblfo_val * voice->viblfo_to_pitch
>                     + voice->modenv_val * voice->modenv_to_pitch) / 
> voice->root_pitch;
> 
> +  printf("");
> +
>    /* Transfer the phase from the voice into the dsp loop parameter
>       dsp_phase */
>    fluid_phase_set(dsp_phase, voice->phase);
> 
> 
> Also you can try change sample rate in fluidsynth.
> Not sure, but imho fluidsynth have not so visible tuning problem. I currently 
> don't have
> enough time, but i really want maximum sound quality and i try fully review 
> and test
>  it later.
> 

There is one issue that can still be a problem and that is that if
FluidSynth is internally configured at a different baud rate than what
the driver is using.  This will usually be reported as a warning though
(perhaps it should actually be a fatal error, at least until drivers can
switch rates on the fly).  That will make all the notes out of tune
though, not just selective few.

The other thing that can cause different sounding audio is with
Instruments with more than one voice sounding at the same time.
FluidSynth currently doesn't start all voices synchronously, so there is
the chance that voices will get started in the next buffer fragment
(multiple of 64 samples by default).  This is also the case with stereo
sample pairs, which is actually contrary to the SF spec.  This will
cause different sounding playback for different note on events (phasing
effects).  This might be what you are hearing, since that printf would
cause some timing differences.

If you discover anything else, be sure to let us know :)
        Josh