If that were entirely true, you would never receive garbled P25, yet it happens. I guess I am going to have to try some experiments for you.
Experiment away, but I'm telling you from a protocol standpoint your hypothesis that random data can be mistaken for AMBE codewords is just not possible. Clearly you shouldn't take my word for it, but go read TIA-102 (as I have done for many hours while working on op25) and understand how the numerous layers of FEC are applied to their respective payloads, then tell me how the firmware in the receiver could even begin to think it has valid data that should be sent for playback.
Garbled playback happens very rarely on a subscriber radio because they have better demodulators and first class FEC handling. At the opposite end of the spectrum, traditional consumer scanners do a horrible job and spit out garbled audio frequently because the do the least well at demodulating the signals and spend little to no effort on FEC handling. Somewhere in the middle are the various SDR packaged (DSD+, OP25, Trunk Recorder,...) and products like the Unication G-series pagers which make an honest effort to get things decoded properly. When garbled audio does work through those devices, it is either the result of the frame having bit errors in the non-error checked portion of the codewords, or attempting to play back encrypted traffic as if it were unencrypted. It is also possible under marginal signal conditions for frames to only successfully being decoded intermittently, leaving the failed frames in between to be either muted or excessively frame-repeated using prior good data.