In response to a target recording from an AM radio source:

The way you record might be bogus. I have tried it too, with variable success. It can't be used in court because of the method. They would claim it to be just "noise" or a channel a bit off if you are able to pick up "voices". The signal can, however, be recorded as ingress ( https://en.wikipedia.org/wiki/Coaxial_cable#Signal_leakage ) into a computer system. I have made thousands of recordings. You can listen to some of them here:

https://soundcloud.com/steven-jones-381108333

To make an ingress recording, start whatever program you record with and then unplug your microphone. Yes, unplug your microphone and any other input source like a webcam.
The resulting file should be blank, but isn't if you are being zapped. All you need to do then is increase the volume. With sound editing software, you can remove hiss, pops/clicks, and other enhancements.

Under spectrographic analysis the sounds can be identified as mechanical productions of what sounds like voices, most of the time. Other times might actually be voices of the people doing it. It is very difficult to get what are called harmonic ladders in these types of recordings. Harmonic ladders are what the courts would use to identify voices and for voice "fingerprinting".

I have suggested that targets use software called GoldWave. There is a demo available, license for only $45 it does a lot! Pop and click removal, initial noise reduction, and most importantly a time function which you can slow down the recording. Much of the sound targets are hearing is transmitted at a high rate of speed, that is, the "radio" signal is sped up before transmission to a target. Upon examination of some recordings, separate and distinct sounds as "voices" can be heard at various rates of playback. The spectrographic display feature in Goldwave is really good for targets who wish to identify actual voices in their recordings, even if the words aren't understood.

Best of luck recording.