If this keeps up, I'd recommend starting with the scientific approach. First would be to purchase a highly sensitive microphone (obviously) with a battery powered recorder (for portability to walk around in the directions of the sounds). The recorder would need a very-very-very clean power supply to keep the noise floor low, since the voices would be in the noise floor and hard to pick out. While designs like this aren't very difficult (going back all the way to the 1960's), this tends to make the hardware expensive when it shouldn't.
The second approach, which should be run right beside the first, is to also add a second recorder but swap the microphone out with a pickup coil. The coil could be an off the shelf air core inductor, either purchased or hand wound with magnet wire on a plastic bobbin. The inductor would pick up any electromagnetic or RF noise that the brain could pick up and interpret as voices or sounds. The diameter of the coil and length of wire will dictate which frequencies it will more easily pick up, and that can get into some really complicated RF antenna math, but keep it simple starting out and just get something working.
I realize y'all have that big hospital bill coming, so this may not be very cost effective right now. What this boils down to is "ghost hunting" hardware. If you buy hardware specifically marketed as ghost hunting, you will get something that is over priced and under delivers.
If you, or someone you know, has experience with hand soldering electronics, the more expensive low noise amplifiers and power supplies aren't too hard to build using cheap components (just slow and tedious). Then the amplified signals could be sent at full volume into a cheaper recorder.
