This may be the case at present but I expect that the various tracks will be digitally filtered in due course. So that actual voices can be interpreted.
Fortunately there is already ready-made sophisticated audio spectral editing software for this, e.g, Izotope RX4. It easily allows a skilled forensic analyst to null out specific spectral patterns such as alarms which obscure speech.
Are these statements still true for modern, digital CVR? For the analogue recording I believe that you can filter a lot, because "it is all there" but buried, in a digital recording there is nothing between the bits. It is a bit like all the higher frequency events not recorded on the FDR with 1 second sampling rate. An oscillation at 1.1 Hz is recorded as an oscillation at .1 Hz as the real frequency is masked by the sampling rate.
What is the sampling rate for digital CVR? How does it match with computer generated sounds (alarms, chimes, cavallery fanfare...) which are also having their own sampling rate or rectangular waveforms. I never worked with digital audio, do you also have some frequency changes similar to my FRD example, so a sound of 42 kHz sampled at 41 kHz is recorded as a 1 kHz sound?