Re: [fluid-dev] Possible bugs in mod and release envelope math

[David Henningsson]

On 2011-07-10 02:07, Andrew Suffield wrote:
> I've been working through the source, and while I don't pretend to
> fully understand what's going on here, I've spotted a couple of things
> that seem wrong:
>
>
> fluid_mod_get_value sets range1 and range2 to 127, which are used as
> the divisor for linear and bipolar modulators. That means, for a
> bipolar mod, an input of 0 maps to -1 and an input of 127 maps to 1,
> and there is no input which maps to 0. I looked at SF2.01 and it says
> the range should be from -1 to 127/128 (ie, range1 and range2 should
> be 128).
>
> This has the unfortunate consequence that the default pan setting of
> 64 does not generate identical output on both channels (which is how I
> spotted this), and the optimisation in fluid_rvoice_buffers_mix is
> defeated because there will never be a centered stereo sample
> (although since it's clearly never worked, the merit of that
> optimisation is questionable).

Thanks for the observation. But just changing the divisor to 128 seems
to me like that would make us lose the right endpoint (as 127/128 < 1)
instead, i e there would be no way to pan hard right. While diverging
from the standard is a bad idea in general, I wonder if there is a point
in trying to do that in this case?

An formula Y = (X-1)/126 would give both 0 and 1 as hard left, 64 as
center and 127 as hard right. Does that make sense to everybody?

> fluid_voice_update_param contains these:
>
>    case GEN_VOLENVRELEASE:             /* SF2.01 section 8.1.3 # 38 */
>      x = _GEN(voice, GEN_VOLENVRELEASE);
>      fluid_clip(x, FLUID_MIN_VOLENVRELEASE, 8000.0f);
>      count = 1 + NUM_BUFFERS_RELEASE(x);
>      fluid_voice_update_volenv(voice, FLUID_VOICE_ENVRELEASE,
>                              count, 1.0f, count ? -1.0f / count : 0.0f, 0.0f, 
> 1.0f);
>      break;
>
>    case GEN_MODENVRELEASE:                                  /* SF 2.01 section 
> 8.1.3 # 30 */
>      x = _GEN(voice, GEN_MODENVRELEASE);
>      fluid_clip(x, -12000.0f, 8000.0f);
>      count = 1 + NUM_BUFFERS_RELEASE(x);
>      fluid_voice_update_modenv(voice, FLUID_VOICE_ENVRELEASE,
>                              count, 1.0f, count ? -1.0f / count : 0.0f, 0.0f, 
> 2.0f);
>
>      break;
>
> Pay particular attention to the 'max' parameter on the release
> section. It's set to 1 for volenv and 2 for modenv. Both volenv and
> modenv have a max of 2 for the decay and sustain sections, so there's
> no apparent reason for this variation. I'm not really sure what's
> going on here but it feels strange.

I agree that it feels strange, but question is if it matters? For
release volumes, things should always ramp down anyway so any value >= 1 
should be okay.

// David