The way to understand this is to break down the terminology. First one to look at is "Modulation" which is the term used to describe the encoding of a "carrier" signal with another "source" signal.
In the case of "amplitude modulation", the "amplitude" of a carrier wave is "modulated" with a source signal in a transmitting circuit. The "receiving" circuit is able to "demodulate" the carrier to recover the "source" signal.
The same happens with "frequency modulation", where the "frequency" of the carrier wave is "modulated" with the source signal and the overall result is then "demodulated" to recover the source signal in the receiver.
FM is the acronym used in radio to describe the use of "Frequency Modulation".
So let us look at the use of "FM Synthesis" which is basically using similar modulation techniques to produce new sounds. Whereas in radio, you are unable to listen to the modulated carrier wave, which exists in the many Megahertz of RF frequency, with FM synthesis all the resulting sounds are all in the audible range, between 20Hz and approximately 16-18KHz, depending on hearing and age.
In order to get an FM Synthesis card to reproduce a voice, you would have to work backwards from an actual voice sound and work out how to modulate an unknown source signal in order to produce an audible voice - via FM Synthesis. Your mission, should you choose to accept it, would be to work out what the unknown source signal is and the FM Modulation parameters that are required to end up with a synthesised voice.
Voice is made up of a combination of tonal sounds and atonal/noise sounds. There are endless papers on the types of sounds that make up voice, which are too numerous to mention here, but that can be easily googled. It is definitely possible to produce both the main types of sounds using an FM synthesiser, however I would suggest it is tricky to produce them all at the same time, which would be required in this application.
You would also be advised to research the CELP codec, which by design breaks down a voice sound into its component parts and then encodes it for transmission to a similar codec which can re-constitute the voice.