It is not direct modulation of the light frequency, but more like this:
200 kHz squarewave VCO + audio on DC pedestal as the control voltage > 200 kHz +/- 0 to 50 kHz squarewave > IR emitter driver
That makes more sense to me than AM plus FM modulation.
So it's square-wave FM modulation.
One easy way to generate this FM signal is to use the VCO section of a CD4046 phase-locked-loop.
---------------
Below is another option which uses a simple PWM circuit to modulate the LED.
It consists of a CMOS relaxation oscillator whose pulse width (and somewhat the frequency) is modulated by the input audio signal.
The output PWM varies from about 50kHz to 70kHz with the component values shown.
The audio is easy to recover with a simple low-pass filter as shown in the bottom right of the schematic, which is a 3rd order, Sallen-Key active filter with a 10kHz corner.
The PWM input would normally be the output of the photo-detector circuit (not shown).
The simulation is shown for input frequencies of 1kHz, 4kHz, and 8Khz.
The linearity is more than adequate for voice, with the input to output distortion being about 1%.
The output has a slight 0.5dB peak at 8kHz due to the 1dB ripple Chebyshev filter.