communication - When to process beam forming algorithms?

I am building a system that uses beam forming to locate noises at different spots. All of the spots will be in front of the microphones (180 degree range of spots).

My first question is if the beam forming algorithms have to be done when receiving the data from the microphones or is it possible to store the information raw and apply algorithms for beam forming after. I ask this because there is multiple locations at random times so we would have to modify the beam forming on the spot which would require faster, more expensive hardware. If we can store the data then apply algorithms we can save cost because we won't need as much speed.

Is identifying multiple noise locations at random even possible?

I am looking for specific sounds, so other sounds will need to be cancelled out or be able to be ignored. This is why I can't use a simple technique as where did the loudest sound came from.

Thank you in advance!