In order to de-reberberate properly, you should know the impulse response (or frequency domain equivalent) of the room, measured from the source to the microphone and then perform de-convolution. Otherwise you are tackling the much more difficult task of blind deconvolution.

If the source is spread out and not a single point then there is no single impulse response, and blind deconvolution would be the only way. If you make certain reasonably valid assumptions about the nature of the impulse response, then you may get a degree of de-reverb which is acceptable, but try too hard and you will introduce artefacts. It's a bit like noise reduction or sharpening images in that respect.

Another way to go would be to convert the signal into many frequency bands and apply a noise gate to attenuate frequency components below a certain level. This would work best with sounds that had fairly abrupt endings. Once again you may get a modest improvement before unpleasant artefacts get introduced.

Above all, the proof of the pudding is in the eating. I look forward to the trial.