[ardour-dev] Deafening noise from plugins when auditioning/unmuting

[Tom Szilagyi]

On Thu, 25 Aug 2005, Per Sigmond wrote:

> Hi.
> I have this problem in the following curcumstances:
[...]

I made some additional tests this evening. Hacked together a
simple 2in/2out 'disk writer' LADSPA plugin that does nothing
to its input, just outputs it and writes it to a 32bit float
WAV file as well (using libsndfile).

I placed this plugin *before* the reverb on the bus which
caused problems, and by viewing the resulting file in audacity,
found that the feedback problem was also present at the input
of the bus. Which means, somehow a 'closed loop' has been
established for a short time in the jack graph. ATM I can find
no other explanation as to how could the output of the bus
(routed only to the master bus) come back and appear at the
input of the bus. This of course causes an enormous signal in
no time when the reverb is applied to it.

I have noted that this whole issue did not occur when I have
set up a simple ardour session for exactly this purpose (just
1 stereo track with audio region, routed to bus with reverb on
it, routed to master bus). I had to try it in one of my
'production' sessions which currently has about 15 tracks.

I have also noticed a phenomenon previously that may be related
to this issue. Sometimes Ardour gets kicked from jack (in fact,
this almost always happens at the end of a recording session when
playback & recording is being stopped, never in the middle of the
recording or when there is only playback). On these occasions I
restart Ardour, and while the session is loading, I sometimes hear
a very short sound impulse in which I can recognize some of the
previously recorded tones, so it's not just random noise, but
some existing session data.

I imagine that something similar happens when the jack graph has
reached a certain complexity and the transport is stopped. This
impulse is harmless in itself, but it seems that it drives the
reverb crazy. The fact of feedback occurring is still unexplained
to me, though. (I have a vague idea that the feedback sound only
occurs with the reverb because there is a one-buffer delay in the
loop which would defeat the positive feedback if no reverb was there.
Hovewer, since the reverb produces output that does not end when the
input ends but lasts longer, it is possible that a positive feedback
loop is established.)

Trivia:
  Ardour 0.600.5 with libardour 0.896.4;
  jackd 0.100.0 with a period size of 1024, 2 periods;
  linux-2.6.13-rc6-rt13 (Ingo's RT preemption patch),
all compiled from source on Debian Sarge.


Tom