A spectrogram always has some non-zero amplitude value at all frequencies, even when it seems there is nothing there (values like 5.1e-007). Two main properties determine what frequencies are "seeable". First is sampling rate: a recording made at 48K has values up to 24K and absolutely nothing above, and one made at 22K maxes out at 11K. This is a fundamental, baked-in definitional limit known as the Nyquist frequency. As a sub-case of tha sampling rate matter, it is possible to manipulate a sound file to change the sampling rate so that a sample made at 22K has no information above 11K, but you can do things to the filee to create a 48K sampling rate file, which has "misinformation" above the original sampling rate. First thing, check the sampling rates.
Second, check the microphone and other front-end frequency characteristics. Various headset microphones especially are designed to maximize speech characteristics so they will filter out anything above 9K (for instance), and may also filter out anything below 100Hz. I've made two-microphone recordings of a single speech event and can compare the frequency characteristics of microphones, and found a striking "dead silence" at 9K with a Zoom-oriented headphone microphone with noise cancelling and some kind of "clarification" technology, compared to a standard "record everything" microphone.
In you look at the spectral slice, you will probably find that the "silent" zone is not exactly 9K, it's e.g. 8686.19. This is controlled by the "dynamic range", and changing from 60dB to 100dB will reveal a bunch of stuff that you could not see before. The cutoff is, more accurately, "around 9K".