It's narrow FM with a "voice scrambler". Before getting passed to the FM modulator the voice signal has its frequency components "inverted", in this case by being subtracted from a frequency of about 3,330 Hz. Some radios (not amateur ones, but then again this isn't the amateur band) use this as a sort of basic privacy feature, but it doesn't do much.
You can get clear voice by passing the output of the FM demodulator to the input of an lower sideband demodulator with a carrier frequency of 3330 Hz and a passband of 300 - 3000 Hz (give or take). This may or may not be directly possible with SDR#. I was able to do it with gqrx by running two copies of the app — one using the IQ file source and outputting to a PulseAudio "null sink", and the second one using the SoapySDR Audio source connected to the output of the first one.