hi hans,
there is a simple method for feedback stabilization coming from adaptive line enhancement theory using FIR N-LMS algorithms. i've checked in my adaptive filter library into CVS today (externals/iem/iem_adaptive); so it should be easy to implement ... back to your problem:
the basic scheme of an adaptive line enhancer with FIR structure is depicted in the figures attached. due to short term periodic components within the target signal, the filter tries to suppress also parts of the target signal. with a small stepsize parameter mu={0.01, 0.00001} this behaviour can be avoided and only long term periodic components of the signal will be suppressed; this also guaranties filter stability. because of the small step size the filter converges slowly and feedback distortions will be attenuated after a nonnegligible period of time - so it will be possible to "play" with feedbacked sounds. the main parameter is the step size. the delay ND considers the short term stationarity of the target signal (eg speech ~20ms). you may also use a feedbacked FIR structures (cp IIR) that provide sharper notches, i.e. less distortion of your target signal. using IIR may become unstable!!
i used this algorithm for residual feedback suppression in communication setups and it worked quite well.
cheers
markus