I think one avenue almost feasible today would be: Electromyography on the larynx > silent speech recognition > transmission of the data through wireless netwo

I think one avenue almost feasible today would be: Electromyography on the larynx > silent speech recognition > transmission of the data through wireless network > speech synthesis > ouput audio to wireless earphone. If you implant both ends you have something indistinguishable from (consensual) telepathy.