I'm no expert on paranormal phenomena by any means, but I do know a bit about electronics, and I also believe that spirits manifest themselves as, and have the ability to manipulate, electromagnetic fields. There are many accounts of this, such as unplugged lights and TVs operating, phones ringing, and voices recorded onto tapes where no sound was heard during the recording.
In reality, recorders don't record sound, they record electromagnetic signals. They contain devices to convert sound into electricity (microphone), and another device to convert the electricity back into sound (speaker), but the guts of the recorder are acting on electrical signals, not sound waves. These electrical signals are then stored on a tape by converting them to a magnetic field, or converted into digital ones and zeros and written onto a memory chip, disc, or tape (in the case of digital recorders).
I believe when spirits "speak" using EVP, they are manipulating the electromagnetic field around the recorder in such a way that the electronics pick up the signal (which is in essence interference) and record it on the medium. The microphone doesn't come into play at all, unless it's acting as an "antenna".
It's similar to how radios pick up signals from the air. In fact, if you've ever been "lucky" enough to live near an AM radio tower, or near some punk with a boosted CB radio, you've probably heard their voices come in over your stereo, TV, telephone, etc. It's the same idea. The electromagnetic waves from the transmitting tower interact with the circuits in the device, which then amplify and convert it into sound. Spirits do the same thing, but on a smaller scale. They are like tiny radio stations.
Some people think that the spirit somehow impresses their voice directly onto a tape. I don't believe it works this way, since you still have to be recording to get anything, and since EVP also works with digital recorders, I doubt a spirit would be able to manipulate digital data to impress a voice on a tape. More likely they're doing it at the analog stage, interfering with the mic preamp or something, which then gets their voice recorded.