Dealing with missing and damaged features (and in particular template and weight adjusting) is the key to making effective use of "auditory scene analysis" processing, blind separation, etc..
No amount of processing can restore information that has been lost. If we know the part that is lost, then we must ignore it. If we know that it has been modified, then we must adjust for this modification.
The role of auditory scene analysis or interference suppression modules is not just to remove interference. It is to provide a map of the reliability of what is left over.
The effectiveness of these principles has been demonstrated in simple experiments (de Cheveigné, 1993). Incorporating them into a real SR system is a challenge, that is well worthwhile. It is a key to the use of ASA, blind separation, and possibly audio-visual cues.
A first thing to do is to design a recognition system with the proper hooks.
A second thing to do is choose a simple task to demonstrate the effectiveness of these schemes. Concurrent voice suppression is on such task, but there are certainly others.