An idea from @seankross :
Move all the Coqui TTS functionality into a separate R package, create a Docker image with Coqui TTS installed, and call
tts_coqui() on the Docker image. Not sure if I described this correctly, but that's the general idea.
The R package would be called roqui 😄